AdCP · Ecosystem Tracker

Who has actually
shipped an agent.

The public agenticadvertising.org registry, tracked as a dated record rather than a snapshot in a slide. Observed automatically by the signals adaptor's watcher, which polls the registry daily at 13:00 UTC; this page is refreshed from its state snapshots when something changes. Last observed July 28, 2026.

21 registered agents

16 sales
3 signals
1 creative
1 buying

Only 7 of 21 answer tools/list publicly →

The registry, agent by agent

	Agent	Type	Discovery	Endpoint
	AdCP Test Agent	Sales	Key-gated	test-agent.adcontextprotocol.org
A seller-side agent: exposes a publisher or SSP’s inventory to buying agents — discovery, availability, and deal construction over MCP instead of a UI or a bespoke API integration. MCP endpoint `https://test-agent.adcontextprotocol.org/mcp` Tools — probed 2026-07-22 · Discovery itself requires credentials — the agent exists, but you need a key even to list its tools. HTTP 401 — bring a key. Probe it yourself (standard MCP discovery) `curl -X POST "https://test-agent.adcontextprotocol.org/mcp" \ -H "content-type: application/json" \ -H "accept: application/json, text/event-stream" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'`
	Advertible Inc.	Sales	Open · 2	adcp.4dvertible.com
A seller-side agent: exposes a publisher or SSP’s inventory to buying agents — discovery, availability, and deal construction over MCP instead of a UI or a bespoke API integration. MCP endpoint `https://adcp.4dvertible.com/mcp` Tools — probed 2026-07-22 · tools/list answered publicly — the catalog below is the live capture. `list_creative_formats`List Available Formats `preview_creative`Generate preview renderings of a creative manifest. For bulk requests (e.g. 'create many ads with size 300x250 in an amazon-like style'), use the settings.size.w/h for the ad size in pixels and settings.brand_type for the style keyword. Probe it yourself (standard MCP discovery) `curl -X POST "https://adcp.4dvertible.com/mcp" \ -H "content-type: application/json" \ -H "accept: application/json, text/event-stream" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'`
	Adzymic (Mediacorp)	Sales	Error	mediacorp.sales-agent.adzymic.ai
A seller-side agent: exposes a publisher or SSP’s inventory to buying agents — discovery, availability, and deal construction over MCP instead of a UI or a bespoke API integration. MCP endpoint `https://mediacorp.sales-agent.adzymic.ai/mcp` Tools — probed 2026-07-22 · The endpoint answered, but not with a valid MCP tools/list response. Bad Request: Missing session ID (HTTP 400) Probe it yourself (standard MCP discovery) `curl -X POST "https://mediacorp.sales-agent.adzymic.ai/mcp" \ -H "content-type: application/json" \ -H "accept: application/json, text/event-stream" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'`
	Adzymic (SPH)	Sales	Error	sph.sales-agent.adzymic.ai
A seller-side agent: exposes a publisher or SSP’s inventory to buying agents — discovery, availability, and deal construction over MCP instead of a UI or a bespoke API integration. MCP endpoint `https://sph.sales-agent.adzymic.ai/mcp` Tools — probed 2026-07-22 · The endpoint answered, but not with a valid MCP tools/list response. Bad Request: Missing session ID (HTTP 400) Probe it yourself (standard MCP discovery) `curl -X POST "https://sph.sales-agent.adzymic.ai/mcp" \ -H "content-type: application/json" \ -H "accept: application/json, text/event-stream" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'`
	Affinity Answers MCP Agent	Signals	Open · 2	mcp.affinityanswers.com
A signals provider: audience and contextual signals as agent-callable tools — capability discovery, natural-language signal search, and activation to a destination platform, with data access key-gated. MCP endpoint `https://mcp.affinityanswers.com/mcp` Tools — probed 2026-07-22 · tools/list answered publicly — the catalog below is the live capture. `get_adcp_capabilities`Get AdCP capabilities supported by this signals agent. Call this first to discover what protocols and features this agent supports. `get_signals`Get signals from the AA first-party signal catalog. Supports two modes: - Discovery: pass signal_spec (natural-language description of desired signals) - Lookup: pass signal_ids (list of {provider, id} objects for exact retrieval) Both modes can be combined: signal_ids anchor the starting set, signal_spec guides further filtering. Probe it yourself (standard MCP discovery) `curl -X POST "https://mcp.affinityanswers.com/mcp" \ -H "content-type: application/json" \ -H "accept: application/json, text/event-stream" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'`
	BidMachine Seller Agent	Sales	Key-gated	adcp.bidmachine.io
A seller-side agent: exposes a publisher or SSP’s inventory to buying agents — discovery, availability, and deal construction over MCP instead of a UI or a bespoke API integration. MCP endpoint `https://adcp.bidmachine.io/adcp/mcp` Tools — probed 2026-07-22 · Discovery itself requires credentials — the agent exists, but you need a key even to list its tools. HTTP 401 — bring a key. Probe it yourself (standard MCP discovery) `curl -X POST "https://adcp.bidmachine.io/adcp/mcp" \ -H "content-type: application/json" \ -H "accept: application/json, text/event-stream" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'`
	Celtra	Creative	Error	adcp-mcp.celtra.com
A creative agent: creative assets and generation/validation workflows exposed as tools, so a buying agent can source or adapt creative inside the same protocol it transacts in. MCP endpoint `https://adcp-mcp.celtra.com/mcp` Tools — probed 2026-07-22 · The endpoint answered, but not with a valid MCP tools/list response. Bad Request: No valid session ID provided (HTTP 400) Probe it yourself (standard MCP discovery) `curl -X POST "https://adcp-mcp.celtra.com/mcp" \ -H "content-type: application/json" \ -H "accept: application/json, text/event-stream" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'`
	Content Ignite	Sales	Error	sales-agent.contentignite.com
A seller-side agent: exposes a publisher or SSP’s inventory to buying agents — discovery, availability, and deal construction over MCP instead of a UI or a bespoke API integration. MCP endpoint `https://sales-agent.contentignite.com` Tools — probed 2026-07-22 · The endpoint answered, but not with a valid MCP tools/list response. non-MCP or unexpected response (HTTP 500) Probe it yourself (standard MCP discovery) `curl -X POST "https://sales-agent.contentignite.com" \ -H "content-type: application/json" \ -H "accept: application/json, text/event-stream" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'`
	Cora AI Sales Agent	Sales	Open · 24	sales-agent.coraai.org
A seller-side agent: exposes a publisher or SSP’s inventory to buying agents — discovery, availability, and deal construction over MCP instead of a UI or a bespoke API integration. MCP endpoint `https://sales-agent.coraai.org/mcp` Tools — probed 2026-07-22 · tools/list answered publicly — the catalog below is the live capture. `get_adcp_capabilities`Discover supported AdCP tasks, seller portfolio, auth, and runtime capabilities. `list_accounts`List buyer accounts available to this credential. `sync_accounts`Provision or update sandbox buyer account relationships for AAO compliance tests. `sync_governance`Persist buyer governance agent references for sandbox account tests. `get_products`Discover sellable products for a buyer brief. `create_media_buy`Create a media buy request against selected Cora AI products. `update_media_buy`Update a previously submitted media buy. `get_media_buys`Return one or more media buys created by this Cora AI test agent. `get_media_buy_delivery`Return delivery metrics for a media buy. `list_creative_formats`Return creative format definitions referenced by CORA AI products. `sync_creatives`Accept sandbox creative assets and attach them to pending media buys. `list_creatives`Protected authentication probe and minimal creative library listing. `validate_input`Dry-run validation for AdCP 3.1 canonical creative manifests. `create_property_list`Create a sandbox property list for governance and targeting tests. `list_property_lists`List sandbox property lists with cursor pagination. `get_property_list`Return full details for a sandbox property list. `update_property_list`Replace metadata or base properties on a sandbox property list. `delete_property_list`Delete a sandbox property list. `validate_property_delivery`Validate delivery records against a sandbox property list. `create_collection_list`Create a sandbox collection list for governance and targeting tests. `list_collection_lists`List sandbox collection lists with cursor pagination. `get_signals`Discover public Cora AI contextual signals. `activate_signal`Activate or deactivate a sandbox Cora AI signal on a platform or sales agent destination. `comply_test_controller`Sandbox-only compliance controller for deterministic AAO storyboard seeding and state forcing. Probe it yourself (standard MCP discovery) `curl -X POST "https://sales-agent.coraai.org/mcp" \ -H "content-type: application/json" \ -H "accept: application/json, text/event-stream" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'`
	Dstillery	Sales	Error	adcp-signals-agent.dstillery.com
A seller-side agent: exposes a publisher or SSP’s inventory to buying agents — discovery, availability, and deal construction over MCP instead of a UI or a bespoke API integration. MCP endpoint `https://adcp-signals-agent.dstillery.com/mcp` Tools — probed 2026-07-22 · The endpoint answered, but not with a valid MCP tools/list response. Bad Request: Missing session ID (HTTP 400) Probe it yourself (standard MCP discovery) `curl -X POST "https://adcp-signals-agent.dstillery.com/mcp" \ -H "content-type: application/json" \ -H "accept: application/json, text/event-stream" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'`
	Equativ	Sales	Open · 1	adcp.equativ.com
A seller-side agent: exposes a publisher or SSP’s inventory to buying agents — discovery, availability, and deal construction over MCP instead of a UI or a bespoke API integration. MCP endpoint `https://adcp.equativ.com/v1/discover` Tools — probed 2026-07-22 · tools/list answered publicly — the catalog below is the live capture. `get_products`Get a listing of current deals relevant to your requirements. Probe it yourself (standard MCP discovery) `curl -X POST "https://adcp.equativ.com/v1/discover" \ -H "content-type: application/json" \ -H "accept: application/json, text/event-stream" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'`
	InMobi Exchange ADCP NonProd	Sales	Key-gated	adcp-nonprod.exchange.inmobi.com
A seller-side agent: exposes a publisher or SSP’s inventory to buying agents — discovery, availability, and deal construction over MCP instead of a UI or a bespoke API integration. MCP endpoint `https://adcp-nonprod.exchange.inmobi.com/mcp` Tools — probed 2026-07-22 · Discovery itself requires credentials — the agent exists, but you need a key even to list its tools. HTTP 401 — bring a key. Probe it yourself (standard MCP discovery) `curl -X POST "https://adcp-nonprod.exchange.inmobi.com/mcp" \ -H "content-type: application/json" \ -H "accept: application/json, text/event-stream" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'`
	InMobi Sales Agent	Sales	Key-gated	adcp-prod.exchange.inmobi.com
A seller-side agent: exposes a publisher or SSP’s inventory to buying agents — discovery, availability, and deal construction over MCP instead of a UI or a bespoke API integration. MCP endpoint `https://adcp-prod.exchange.inmobi.com/mcp` Tools — probed 2026-07-22 · Discovery itself requires credentials — the agent exists, but you need a key even to list its tools. HTTP 401 — bring a key. Probe it yourself (standard MCP discovery) `curl -X POST "https://adcp-prod.exchange.inmobi.com/mcp" \ -H "content-type: application/json" \ -H "accept: application/json, text/event-stream" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'`
	LoopMe Sales Agent	Sales	Key-gated	platform.loopme.ai
A seller-side agent: exposes a publisher or SSP’s inventory to buying agents — discovery, availability, and deal construction over MCP instead of a UI or a bespoke API integration. MCP endpoint `https://platform.loopme.ai/mcp/seller` Tools — probed 2026-07-22 · Discovery itself requires credentials — the agent exists, but you need a key even to list its tools. HTTP 401 — bring a key. Probe it yourself (standard MCP discovery) `curl -X POST "https://platform.loopme.ai/mcp/seller" \ -H "content-type: application/json" \ -H "accept: application/json, text/event-stream" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'`
	No Fluff Advisory — Evgeny (Ev) Popovthis site	Signals	Open · 8	adcp.signal-stack.io
A signals provider: audience and contextual signals as agent-callable tools — capability discovery, natural-language signal search, and activation to a destination platform, with data access key-gated. MCP endpoint `https://adcp.signal-stack.io/mcp` Tools — probed 2026-07-22 · tools/list answered publicly — the catalog below is the live capture. `get_adcp_capabilities`Returns the capabilities of this AdCP Signals Provider: supported operations, signal categories, available destinations, activation mode, and limits. Call this first to understand what the provider supports before searching or activating signals. Optionally pass `protocols` to filter the response to specific protocol blocks. `get_signals`Discover and propose audience signals from this AdCP provider. Pass a natural language signal_spec to get AI-generated custom segment proposals alongside catalog signals. Proposals have signal_type='custom' and is_live=false — activate them to create and deploy. deliver_to is required: specifies target deployments and countries. `activate_signal`Activate an audience signal to a destination platform. Returns task_id immediately with status 'pending' — this is an async operation. Poll get_operation_status with task_id to check progress. Optionally provide webhook_url to receive a POST callback on completion. If activating a custom proposal from get_signals (is_live=false), the segment is created on activation. `get_operation_status`Poll the status of an async activation task. Returns current status: pending → working → completed (or failed). On completion, includes the activated deployments with activation keys. Use the task_id returned by activate_signal. `get_similar_signals`Find audience signals semantically similar to a reference signal using UCP embedding cosine similarity. Returns ranked signals with similarity scores. Demonstrates UCP vector search vs keyword search — related signals cluster by taxonomy, category, and data source quality. `query_signals_nl`Find audience signals matching a natural language description. Decomposes the query into a boolean AST (AND/OR/NOT), resolves each dimension against the signal catalog using hybrid rule+embedding+lexical matching, and returns ranked matches with a compositional audience size estimate. Use when the user describes a target audience in free-form language. `get_concept`Look up a specific concept from the UCP Concept-Level VAC registry by concept_id. Returns the concept definition including label, description, constituent dimensions, and cross-taxonomy member nodes showing how it maps to IAB, LiveRamp, TradeDesk, etc. `search_concepts`Semantic search over the UCP Concept Registry. Find concepts matching a natural language description. Use this to discover concept_ids for use in audience queries, or to explore what cross-taxonomy mappings exist for a given audience type. Optionally filter by category: demographic, interest, behavioral, geo, archetype, content, purchase_intent. Probe it yourself (standard MCP discovery) `curl -X POST "https://adcp.signal-stack.io/mcp" \ -H "content-type: application/json" \ -H "accept: application/json, text/event-stream" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'`
	Pubx Sales Agent	Sales	Error	selleragent.pubx.ai
A seller-side agent: exposes a publisher or SSP’s inventory to buying agents — discovery, availability, and deal construction over MCP instead of a UI or a bespoke API integration. MCP endpoint `https://selleragent.pubx.ai/mcp` Tools — probed 2026-07-22 · The endpoint answered, but not with a valid MCP tools/list response. No catalog available. Probe it yourself (standard MCP discovery) `curl -X POST "https://selleragent.pubx.ai/mcp" \ -H "content-type: application/json" \ -H "accept: application/json, text/event-stream" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'`
	Purrsonality Seller	Sales	Key-gated	seller.purrsonality.rocketscience.pl
A seller-side agent: exposes a publisher or SSP’s inventory to buying agents — discovery, availability, and deal construction over MCP instead of a UI or a bespoke API integration. MCP endpoint `https://seller.purrsonality.rocketscience.pl/mcp` Tools — probed 2026-07-22 · Discovery itself requires credentials — the agent exists, but you need a key even to list its tools. HTTP 401 — bring a key. Probe it yourself (standard MCP discovery) `curl -X POST "https://seller.purrsonality.rocketscience.pl/mcp" \ -H "content-type: application/json" \ -H "accept: application/json, text/event-stream" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'`
	Purrsonality Signals	Signals	Key-gated	signals.purrsonality.rocketscience.pl
A signals provider: audience and contextual signals as agent-callable tools — capability discovery, natural-language signal search, and activation to a destination platform, with data access key-gated. MCP endpoint `https://signals.purrsonality.rocketscience.pl/mcp` Tools — probed 2026-07-22 · Discovery itself requires credentials — the agent exists, but you need a key even to list its tools. HTTP 401 — bring a key. Probe it yourself (standard MCP discovery) `curl -X POST "https://signals.purrsonality.rocketscience.pl/mcp" \ -H "content-type: application/json" \ -H "accept: application/json, text/event-stream" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'`
	Rediads Private Limited	Sales	Down	adcp.rediads.com
A seller-side agent: exposes a publisher or SSP’s inventory to buying agents — discovery, availability, and deal construction over MCP instead of a UI or a bespoke API integration. MCP endpoint `https://adcp.rediads.com/mcp` Tools — probed 2026-07-22 · The endpoint did not respond within the probe window. This operation was aborted (HTTP —) Probe it yourself (standard MCP discovery) `curl -X POST "https://adcp.rediads.com/mcp" \ -H "content-type: application/json" \ -H "accept: application/json, text/event-stream" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'`
	Swivel	Buying	Open · 11	adcp.swivel.ai
A buyer-side agent: represents an advertiser or platform in discovery and transaction — the counterpart that calls sales, signals, and creative agents to assemble a buy. MCP endpoint `https://adcp.swivel.ai/mcp` Tools — probed 2026-07-22 · tools/list answered publicly — the catalog below is the live capture. `get_products`Retrieve products based on brief description, promoted offering, and filters. All parameters are optional. Filters must be nested under the "filters" key using their exact field names (e.g., filters.standard_formats_only, filters.delivery_type). Do not abbreviate or rename filter fields. IMPORTANT: Only include parameters you actually need. Do NOT send default or placeholder values (0, false, "", []). Omitting a parameter means "no filter". A format_id only needs agent_url and id — do not add width, height, or duration_ms. `list_creative_formats`Discover supported creative formats with filtering by media type, dimensions, and standards compliance. IMPORTANT: Call this tool BEFORE sync_creatives to get valid format_id objects. The sync_creatives tool requires format_id objects from this response - do not use made-up or example URLs. Each format in the response contains a format_id object with: - agent_url: The publisher's format endpoint URL (required for sync_creatives) - id: The format identifier string A format_id only needs agent_url and id. Do NOT add extra properties like width, height, or duration_ms. Workflow: 1. Call list_creative_formats to discover available formats 2. Select a format that matches your creative type (video, display, etc.) 3. Use the complete format_id object (agent_url + id only) when calling sync_creatives IMPORTANT: All filter parameters are optional. Only include parameters you actually need to filter by. Do NOT send default or placeholder values (0, false, "", []). Omitting a parameter means "no filter" — sending a value means "filter to exactly this". Filter options: - format_ids: Look up specific formats by agent_url + id - type: Filter by media type (audio, video, display, native, dooh, rich_media, universal) - asset_types: Filter by asset type (image, video, html, audio) - max_width/max_height/min_width/min_height: Filter by dimensions (omit to return all sizes) - standard_only: Set to true to only return IAB standard formats (omit otherwise) - name_search: Search formats by name `list_authorized_properties`Discover all properties this agent is authorized to represent, with optional tag filtering `list_creatives`Retrieve creatives based on format, tags, asset type, status, and other filters. IMPORTANT: Valid status values are: 'processing', 'approved', 'rejected', 'pending_approval' Note: There is NO 'active' status. Use 'approved' for creatives that are ready to serve. IMPORTANT: Do NOT add a status filter unless the user explicitly asks to filter by status. By default, return all creatives regardless of status. Supports filtering by: - asset_type: 'video' or 'vast' - status: single status filter (only when explicitly requested) - statuses: array of status values to filter by multiple statuses (only when explicitly requested) Example - List all video creatives: { "filters": { "asset_type": "video" } } Example - List approved creatives (only when user asks for approved): { "filters": { "status": "approved" } } Example - List creatives pending approval or processing: { "filters": { "statuses": ["pending_approval", "processing"] } } Example - List creatives with pagination: { "filters": { "asset_type": "video" }, "pagination": { "limit": 20, "offset": 0 } } `create_media_buy`Create a new media buy with packages, targeting, and budget configuration. ## AI Instructions IMPORTANT: Before calling this tool, follow these steps: 1. Gather required information from user - Ask the user for any missing fields: - end_time: Campaign end date (ISO 8601 format, e.g., "2024-12-31T23:59:59Z") - packages: At least one package with: - product_id: The product to purchase (use get_products tool first if needed) - pricing_option_id: The pricing option for the product - budget: The budget amount for this package 2. Auto-generate and prefill these values (do not ask the user): - buyer_ref: Generate as `mb_<campaign_title>_<YYYYMMDD>` (e.g., "mb_summer_sale_20240115") - package buyer_ref: Generate as `pk_<package_name>_<YYYYMMDD>` (e.g., "pk_display_ads_20240115") - brand_manifest: Always set to empty object `{}` - start_time: Always set to `"asap"` unless the user specifies otherwise 3. Ask for notifications configuration (all of these are optional but ask anyways) - reporting_webhook: URL and authentication for reporting delivery performance - push_notification_config: URL and authentication for task completion notifications 4. Present FULL summary for approval - Before calling this tool, show ALL fields in a clear, editable format: "Here's the media buy I'm about to create: Media Buy Reference: mb_example_20240115 Start Time: ASAP End Time: 2024-01-31T23:59:59Z Package 1: - Package Reference: pk_example_20240115 - Product: [Product Name] - Product ID: product-123 - Pricing Option ID: pricing-456 - Budget: $1,000 Let me know if you'd like to change any of these values, or confirm to proceed." This ensures the user sees ALL fields and can request changes before you call the tool. ALWAYS ask for confirmation after requesting more data or making changes to the request. NEVER retry this tool with guessed values if it fails. If validation fails, ask the user for the correct information. `get_media_buy_delivery`Retrieve comprehensive delivery metrics and performance data for media buys `get_media_buys`Poll media buy status, creative approval state, and available actions for one or more media buys `sync_creatives`Sync creative assets with upsert semantics - supports bulk operations, patch updates, and assignment management. IMPORTANT: Before calling this tool, you MUST first call list_creative_formats to get valid format_ids from the publisher. The format_id field requires a FormatIdentifier object obtained from the list_creative_formats response - do NOT use example.com URLs or made-up format IDs. format_id structure (from list_creative_formats response): { "agent_url": "https://actual-publisher-url.com/...", // Real URL from list_creative_formats "id": "format_id_string", // Format ID from the publisher "width": 1920, // Optional dimensions "height": 1080 } assets field format (IMPORTANT - use object, NOT array): The "assets" field must be an OBJECT keyed by role name, NOT an array. CORRECT: "assets": { "video": { "url": "...", "duration_ms": 30000 } } WRONG: "assets": [{ "video": { "url": "..." } }] WRONG: "assets": [{ "url": "..." }] Supported asset types (these go directly in the assets object): - Video: "assets": { "video": { "url": "https://...", "duration_ms": 30000, "width": 1920, "height": 1080 } } - VAST URL: "assets": { "vast": { "delivery_type": "url", "url": "https://vast.example.com/tag", "vast_version": "4.0" } } - VAST Inline: "assets": { "vast": { "delivery_type": "inline", "content": "<VAST>...</VAST>", "vast_version": "4.0" } } Workflow: 1. Call list_creative_formats to get available formats from the publisher 2. Select a format_id object from the formats response that matches your creative type 3. Call sync_creatives with the actual format_id object (not a string, not example URLs) 4. Optionally assign creatives to packages using the 'assignments' parameter ASSIGNING CREATIVES TO PACKAGES (IMPORTANT): To assign creatives to packages/media buys, use the top-level "assignments" parameter, NOT "placement_ids". - "assignments" (top-level): Maps creative_id → array of package IDs. Use this to link creatives to packages. - "placement_ids" (per-creative): For advanced use only - specifies specific placements within a package. Do NOT use this for basic creative-to-package assignment. CORRECT way to assign creative to a package: { "creatives": [{ "creative_id": "my_creative", ... }], "assignments": { "my_creative": ["package-uuid-from-create-media-buy"] } } WRONG - do not use placement_ids for package assignment: { "creatives": [{ "creative_id": "my_creative", "placement_ids": ["package-uuid"], ... }] } Workflow for creative assignment: 1. Call create_media_buy → get package IDs from response (data.packages[].package_id or data.packages[].buyer_ref) 2. Call sync_creatives with "assignments" parameter: { "creative_id": ["package_id1", "package_id2"] } 3. Both package UUIDs and buyer_refs are supported for package identification Example video creative (assuming format_id from list_creative_formats): { "creatives": [{ "creative_id": "brand_video_30s", "name": "Brand Video 30s", "format_id": { "agent_url": "https://publisher.adcontextprotocol.org/formats", "id": "video_preroll" }, "assets": { "video": { "url": "https://cdn.example.com/video.mp4", "duration_ms": 30000 } } }], "assignments": { "brand_video_30s": ["package-uuid-123"] } } Example VAST creative: { "creatives": [{ "creative_id": "vast_preroll", "name": "VAST Pre-roll", "format_id": { "agent_url": "https://publisher.adcontextprotocol.org/formats", "id": "video_preroll" }, "assets": { "vast": { "delivery_type": "url", "url": "https://vast.example.com/tag" } } }] } `update_media_buy`Update campaign and package settings on an existing media buy. PATCH semantics: only the fields you send are changed; omitted fields stay the same. ## When to use this tool - Pause/resume the entire media buy or individual packages. - Shift schedule (start_time / end_time) or budget BEFORE the buy is approved. - Update targeting_overlay at any point in the lifecycle. - Update creative assignments (creative_ids, creatives, creative_assignments) at any point. ## Identifier — use ONE of these, never both - media_buy_id — publisher's UUID (preferred when you have it). - buyer_ref — your reference string for the buy. The tool translates buyer_ref to the underlying UUID using an exact-match lookup. If your buyer_ref does not uniquely identify one buy you will get an AMBIGUOUS_IDENTIFIER error; pass media_buy_id instead. ## Fields that become IMMUTABLE once the buy is approved Once the media buy is in an approved status (active, paused), the following fields cannot be changed for the lifetime of this media buy: - start_time - end_time - package budget - package impressions - package bid_price Retrying the call will not succeed. Do not retry. If these fields need to differ, create a NEW media buy with the desired values. This immutability is per-buy, not per-field — the same fields on a different buy are still editable pre-approval. Other fields (paused, targeting_overlay, pacing, creative_ids, creatives, creative_assignments) remain mutable after approval. ## targeting_overlay is REPLACED WHOLESALE When you send targeting_overlay, the new object fully replaces the previous one. Sub-fields you omit are wiped. If you want a partial update: 1. GET the media buy's current targeting_overlay first. 2. Merge your changes locally into that object. 3. Send the complete merged object in your update. ## Creative field precedence When multiple creative fields are sent for the same package: - creative_assignments wins over creative_ids (it's the authoritative shape with weights and placement targeting). - creatives are uploaded and added to the library first, then assigned via creative_ids/creative_assignments semantics. - creatives with a format_id belonging to a DIFFERENT seller than this media buy's seller are rejected — the buyer cannot attach cross-seller creatives. ## Error handling On validation failure, the tool returns a flat response with an errors array. Possible codes: UPDATE_NOT_ALLOWED (field immutable for this status), INVALID_STATE (buy is completed/rejected), MEDIA_BUY_NOT_FOUND, PACKAGE_NOT_FOUND, AMBIGUOUS_IDENTIFIER, INVALID_FIELD_VALUE, OPERATION_FAILED. NEVER retry on validation failure. Ask the user for corrected input — the server will not accept the same payload a second time. `tasks_list`List and filter async tasks by status, type, domain, date range, or context identifiers. Supports pagination and sorting per AdCP v2.5.3 tasks/list. `tasks_get`Retrieve a specific async task by ID, optionally including full request/response history. Use to poll the status of any async task (create_media_buy, update_media_buy, sync_creatives, activate_signal, etc.) and retrieve results when complete. Probe it yourself (standard MCP discovery) `curl -X POST "https://adcp.swivel.ai/mcp" \ -H "content-type: application/json" \ -H "accept: application/json, text/event-stream" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'`
	vastlint	Sales	Open · 14	vastlint.org
A seller-side agent: exposes a publisher or SSP’s inventory to buying agents — discovery, availability, and deal construction over MCP instead of a UI or a bespoke API integration. MCP endpoint `https://vastlint.org/mcp` Tools — probed 2026-07-22 · tools/list answered publicly — the catalog below is the live capture. `validate_vast`Validate a VAST XML tag against the IAB VAST 2.0-4.3 specification. Returns all issues found with severity, rule ID, location, and spec reference. A document is valid when errors == 0, regardless of warning or info count. Use wrapper_depth when validating a document inside a wrapper chain. `validate_vast_url`Fetch a VAST tag from a URL and validate it. Handles redirects. Use max_depth to control how deep wrapper chains are followed (default 5, per IAB VAST 4.x recommendation). AI agents typically receive VAST URLs rather than raw XML -- use this tool for that case. `list_rules`List the full catalog of VAST validation rules available in vastlint. Returns rule IDs, default severities, and descriptions. Call this once and cache the result -- the catalog is static. Use rule IDs from this list with explain_rule for full details and fix guidance. `explain_rule`Get full details for a specific VAST validation rule: description, spec reference, severity, what triggers it, and how to fix it. Use rule IDs from list_rules. This is the primary tool for understanding and fixing VAST issues flagged by validate_vast. `fix_vast`Auto-fix a VAST XML tag. Applies all deterministic, safe fixes: HTTP → HTTPS upgrades in all URL-bearing elements, and removal of deprecated attributes. Returns the repaired XML, a list of every fix applied (rule ID + description + path), and any remaining issues that require manual intervention. Always re-validate the returned xml with validate_vast to confirm no errors remain. `get_adcp_capabilities`AdCP protocol discovery. Returns the AdCP version, supported protocols, and governance capabilities of this vastlint agent. Call this first when integrating vastlint into an AdCP creative pipeline — it declares which VAST creative features can be evaluated (spec compliance, HTTPS enforcement, wrapper depth). Part of the Ad Context Protocol (AdCP 3.0) specification. `list_content_standards`AdCP content-standards specialism. Lists content governance standards. When called with an 'account' object, returns buyer-defined standards stored for that brand. Without an account, returns the built-in catalog of VAST-relevant standards. Supports cursor-based pagination via the 'pagination' object. Requires Bearer authentication: Authorization: Bearer <api_key>. `create_content_standards`AdCP content-standards specialism. Creates a new set of content quality policies for a brand. Policies include brand safety rules, imagery quality requirements, and compliance constraints. Returns a standards_id that can be used with calibrate_content and validate_content_delivery. Requires Bearer authentication. `get_content_standards`AdCP content-standards specialism. Retrieves the full details of a specific content standard, including all policies and their enforcement levels. Requires Bearer authentication. `update_content_standards`AdCP content-standards specialism. Updates the policies on an existing content standard set. The supplied policies array replaces the existing policy set. Requires Bearer authentication. `delete_content_standards`AdCP content-standards specialism. Permanently removes a content standard set. Callers relying on the standards_id for calibrate_content or validate_content_delivery should retire those integrations first. Requires Bearer authentication. `calibrate_content`AdCP content-standards specialism. Evaluates a creative artifact against a content standard set before delivery. Returns a per-policy verdict and an overall pass/fail result. Use this as a pre-flight check before launching a campaign. Requires Bearer authentication. `validate_content_delivery`AdCP content-standards specialism. Validates that delivered creatives met the content standards. Accepts delivery records with creative references and returns per-record compliance status plus an overall summary. Requires Bearer authentication. `list_creatives`AdCP creative governance. Lists creatives known to this agent. For a validation-only agent like vastlint, this always returns an empty collection — creatives are validated on demand via validate_vast rather than stored. Requires Bearer authentication. Probe it yourself (standard MCP discovery) `curl -X POST "https://vastlint.org/mcp" \ -H "content-type: application/json" \ -H "accept: application/json, text/event-stream" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'`
No agents match — clear the search or filter.

Registration means a live MCP endpoint speaking AdCP — not an announcement, an integration deck, or a waitlist. Click any row for its tool catalog and a ready-to-run discovery call. The Discovery column is a finding in itself (probed 2026-07-22): of 21 registered agents, only 7 answer tools/list publicly — 7 key-gate discovery itself, and 7 error or don't respond. Registered and discoverable are not yet the same thing.

Opentools/list answered publicly — the catalog below is the live capture.
Key-gatedDiscovery itself requires credentials — the agent exists, but you need a key even to list its tools.
ErrorThe endpoint answered, but not with a valid MCP tools/list response.
DownThe endpoint did not respond within the probe window.

What an agent actually returns

Not mockups — real exchanges, captured 2026-07-22: three against this site's own registered signals agent (discovery is public, capabilities are self-reported and versioned, data is key-gated), and a fourth against an open third-party sales agent, because the seller side of the protocol answers with different vocabulary — products, formats, media buys. New to the protocol itself? Start with the vendor-neutral AdCP guide.

1 tools/list — what can this agent do? public

The standard MCP discovery call. This agent exposes 8 tools — capability lookup, natural-language signal search, activation, status polling, and the UCP concept registry.

{
  "tools": [
    {
      "name": "get_adcp_capabilities",
      "description": "Returns the capabilities of this AdCP Signals Provider: supported operations, signal categories, available destinations, activation mode, and limits. Call this first to understand what the provider supports before searching or activating signals. Optionally pass `protocols` to filter the response to specific protocol blocks."
    },
    {
      "name": "get_signals",
      "description": "Discover and propose audience signals from this AdCP provider. Pass a natural language signal_spec to get AI-generated custom segment proposals alongside catalog signals. Proposals have signal_type='custom' and is_live=false — activate them to create and deploy. deliver_to is required: specifies target deployments and countries."
    },
    {
      "name": "activate_signal",
      "description": "Activate an audience signal to a destination platform. Returns task_id immediately with status 'pending' — this is an async operation. Poll get_operation_status with task_id to check progress. Optionally provide webhook_url to receive a POST callback on completion. If activating a custom proposal from get_signals (is_live=false), the segment is created on activation."
    },
    {
      "name": "get_operation_status",
      "description": "Poll the status of an async activation task. Returns current status: pending → working → completed (or failed). On completion, includes the activated deployments with activation keys. Use the task_id returned by activate_signal."
    },
    {
      "name": "get_similar_signals",
      "description": "Find audience signals semantically similar to a reference signal using UCP embedding cosine similarity. Returns ranked signals with similarity scores. Demonstrates UCP vector search vs keyword search — related signals cluster by taxonomy, category, and data source quality."
    },
    {
      "name": "query_signals_nl",
      "description": "Find audience signals matching a natural language description. Decomposes the query into a boolean AST (AND/OR/NOT), resolves each dimension against the signal catalog using hybrid rule+embedding+lexical matching, and returns ranked matches with a compositional audience size estimate. Use when the user describes a target audience in free-form language."
    },
    {
      "name": "get_concept",
      "description": "Look up a specific concept from the UCP Concept-Level VAC registry by concept_id. Returns the concept definition including label, description, constituent dimensions, and cross-taxonomy member nodes showing how it maps to IAB, LiveRamp, TradeDesk, etc."
    },
    {
      "name": "search_concepts",
      "description": "Semantic search over the UCP Concept Registry. Find concepts matching a natural language description. Use this to discover concept_ids for use in audience queries, or to explore what cross-taxonomy mappings exist for a given audience type. Optionally filter by category: demographic, interest, behavioral, geo, archetype, content, purchase_intent."
    }
  ]
}

2 get_adcp_capabilities — versions, protocols, features public

This is where version compliance lives in practice: the agent self-reports supported_versions (here 3.0 and 3.1), its protocol blocks, signing posture, and experimental features. A buying agent calls this first and adapts.

{
  "status": "completed",
  "adcp_version": "3.1",
  "adcp": {
    "major_versions": [
      3
    ],
    "supported_versions": [
      "3.0",
      "3.1"
    ],
    "idempotency": {
      "supported": true,
      "replay_ttl_seconds": 86400
    }
  },
  "supported_protocols": [
    "signals"
  ],
  "specialisms": [
    "signal-owned"
  ],
  "request_signing": {
    "supported": false,
    "covers_content_digest": "either",
    "required_for": [],
    "warn_for": [],
    "supported_for": []
  },
  "webhook_signing": {
    "supported": false,
    "legacy_hmac_fallback": true
  },
  "identity": {
    "per_principal_key_isolation": false
  },
  "experimental_features": [
    "ucp.embedding_bridge",
    "ucp.concept_registry",
    "ucp.gts_anchors",
    "ucp.projector_simulation",
    "dts.label_v1_2",
    "cross_taxonomy.nine_systems",
    "analytics.query_vector",
    "analytics.semantic_arithmetic",
    "analytics.analogy",
    "analytics.neighborhood",
    "analytics.coverage_gaps",
    "analytics.knn_graph",
    "analytics.lorenz",
    "portfolio.pareto_frontier",
    "portfolio.greedy_marginal_reach",
    "portfolio.info_overlap",
    "portfolio.from_brief",
    "federation.a2a_dstillery",
    "federation.cross_similarity",
    "federation.multi_agent_orchestrator",
    "governance.data_hygiene",
    "governance.audit_log",
    "measurement.reach_forecaster",
    "measurement.lift_forecaster_mock",
    "id_resolution.nine_id_types",
    "temporal.seasonality_profiles",
    "temporal.decay_half_life",
    "temporal.volatility_index"
  ],
  "wholesale_feed_versioning": {
    "supported": true
  },
  "signals": {
    "discovery_modes": [
      "brief",
      "wholesale"
    ],
    "in_flight_max_seconds": 30,
    "data_provider_domains": [
      "evgeny.dev"
    ],
    "features": {
      "catalog_signals": true
    },
    "signal_categories": [
      "demographic",
      "interest",
      "purchase_intent",
      "geo",
      "composite"
    ],
    "dynamic_segment_generation": true,
    "activation_mode": "async",
    "provider": "AdCP Signals Adaptor - Demo Provider (Evgeny)",
    "destinations": [
      {
        "id": "mock_dsp",
        "name": "Mock DSP",
        "type": "dsp",
        "activation_supported": true,
        "stage": "live",
        "vendor": "Internal mock",
        "activation_pattern": "push_async",
        "auth_mechanism": "bearer_token (DEMO_API_KEY)",
        "id_types_accepted": [
          "segment_id",
          "hashed_email",
          "maid_ios",
          "maid_android",
          "uid2",
          "ctv_device"
        ],
        "data_format": "segment_id + platform_segment_id pair",
        "segment_refresh_sla": "< 60 seconds (synchronous in demo)",
        "latency_p50_ms": 120,
        "latency_p99_ms": 450,
        "use_cases": [
          "Prospecting",
          "Retargeting",
          "CTV reach",
          "Measurement passback"
        ],
        "activation_flow": "activate_signal -> ephemeral platform_segment_id -> POST segment membership -> get_operation_status polling",
        "docs_url": "https://github.com/EvgenyAndroid/adcp-signals-adaptor#mock-dsp",
        "onboarding": "No onboarding — always live in demo.",
        "notes": "Reference DSP that always returns completed within ~5 seconds. Safe target for any storyboard run."
      },
      {
        "id": "mock_cleanroom",
        "name": "Mock Clean Room",
        "type": "cleanroom",
        "activation_supported": true,
        "stage": "live",
        "vendor": "Internal mock (simulating Snowflake / Habu / InfoSum / AWS CR shape)",
        "activation_pattern": "share_async",
        "auth_mechanism": "cleanroom_agreement_id + bearer",
        "id_types_accepted": [
          "hashed_email",
          "ramp_id",
          "uid2",
          "graph_id"
        ],
        "data_format": "cleanroom table share (SIGNAL_MEMBERS)",
        "segment_refresh_sla": "< 15 minutes",
        "latency_p50_ms": 800,
        "latency_p99_ms": 3500,
        "use_cases": [
          "1P x 2P match-rate analysis",
          "Campaign overlap",
          "Incrementality lift",
          "No-PII joins"
        ],
        "activation_flow": "activate_signal -> grant table share -> cleanroom provider ingests -> membership available on next scheduled compute",
        "docs_url": "https://github.com/EvgenyAndroid/adcp-signals-adaptor#mock-cleanroom",
        "onboarding": "Mock — grants auto-accepted.",
        "notes": "Models the governance boundary: buyer's 1P data never leaves their clean-room tenant. Returns a stub JOIN matrix."
      },
      {
        "id": "mock_cdp",
        "name": "Mock CDP",
        "type": "cdp",
        "activation_supported": true,
        "stage": "live",
        "vendor": "Internal mock (simulating Segment / mParticle / Tealium / Treasure Data shape)",
        "activation_pattern": "push_sync",
        "auth_mechanism": "write_key",
        "id_types_accepted": [
          "user_id",
          "anonymous_id",
          "hashed_email"
        ],
        "data_format": "Audience membership events (audience_entered / audience_exited)",
        "segment_refresh_sla": "real-time (stream)",
        "latency_p50_ms": 60,
        "latency_p99_ms": 220,
        "use_cases": [
          "Personalization",
          "Owned-channel orchestration",
          "Email / push audience sync"
        ],
        "activation_flow": "activate_signal -> stream audience_entered events keyed by user_id -> CDP fans out to downstream tools",
        "docs_url": "https://github.com/EvgenyAndroid/adcp-signals-adaptor#mock-cdp",
        "onboarding": "Mock — write-key auto-provisioned.",
        "notes": "Best fit when the audience drives owned-channel surfaces (email, push, on-site) vs paid media."
      },
      {
        "id": "mock_measurement",
        "name": "Mock Measurement Platform",
        "type": "measurement",
        "activation_supported": false,
        "stage": "live",
        "vendor": "Internal mock (simulating Nielsen DAR / Kantar / IAS / DV shape)",
        "activation_pattern": "read_only",
        "auth_mechanism": "api_key",
        "id_types_accepted": [
          "segment_id"
        ],
        "data_format": "Lift / reach / frequency report rows",
        "segment_refresh_sla": "T+1 day (batch)",
        "latency_p50_ms": null,
        "latency_p99_ms": null,
        "use_cases": [
          "Brand-lift studies",
          "Campaign reach validation",
          "Third-party verification"
        ],
        "activation_flow": "Not an activation target. Read-only: fetches lift/reach reports tagged to a signal's activations.",
        "docs_url": "https://github.com/EvgenyAndroid/adcp-signals-adaptor#mock-measurement",
        "onboarding": "Mock — reports synthesized per signal.",
        "notes": "activation_supported=false by design. Surfaced so planners see measurement coverage alongside activation."
      },
      {
        "id": "ttd_sandbox",
        "name": "The Trade Desk (sandbox)",
        "type": "dsp",
        "activation_supported": false,
        "stage": "sandbox",
        "vendor": "The Trade Desk",
        "activation_pattern": "push_async",
        "auth_mechanism": "oauth2 (TTD Partner Console)",
        "id_types_accepted": [
          "ttd_id",
          "uid2",
          "hashed_email",
          "maid_ios",
          "maid_android"
        ],
        "data_format": "Third-party data segment CSV upload (TTD DMP format)",
        "segment_refresh_sla": "< 24 hours",
        "latency_p50_ms": null,
        "latency_p99_ms": null,
        "use_cases": [
          "Open-web DSP activation",
          "CTV across TTD's OpenPath inventory",
          "UID2 cookieless targeting"
        ],
        "activation_flow": "activate_signal -> request OAuth consent -> upload segment CSV via TTD Partner API -> segment available in TTD UI once ingested",
        "docs_url": "https://partner.thetradedesk.com/v3/portal/api/doc/ThirdPartyData",
        "onboarding": "Requires TTD Partner account + OAuth app registration. Target Q2 2026.",
        "notes": "Roadmap. When live, will be the default DSP for UID2-centric cookieless plans."
      },
      {
        "id": "dv360_sandbox",
        "name": "DV360 (sandbox)",
        "type": "dsp",
        "activation_supported": false,
        "stage": "sandbox",
        "vendor": "Google Display & Video 360",
        "activation_pattern": "push_async",
        "auth_mechanism": "oauth2 (Google Cloud / DV360 API)",
        "id_types_accepted": [
          "hashed_email",
          "maid_ios",
          "maid_android",
          "publisher_provided_id"
        ],
        "data_format": "Customer Match list upload (Google Ads API)",
        "segment_refresh_sla": "< 12 hours",
        "latency_p50_ms": null,
        "latency_p99_ms": null,
        "use_cases": [
          "YouTube reach",
          "Programmatic display",
          "Customer Match seed audiences"
        ],
        "activation_flow": "activate_signal -> OAuth consent -> hash + upload via Customer Match API -> list available in DV360 after match-rate check",
        "docs_url": "https://developers.google.com/display-video/api/reference/rest",
        "onboarding": "Requires Google Ads account + MCC access + OAuth app. Target Q3 2026.",
        "notes": "Roadmap. When live, will be the default DSP for Google-ecosystem plans."
      }
    ],
    "limits": {
      "max_signals_per_request": 100,
      "max_rules_per_segment": 6,
      "default_max_results": 5
    }
  },
  "ext": {
    "compliance": {
      "spec_version": "adcp_3.0",
      "client_runner": "@adcp/sdk@12.0.0",
      "last_run": "2026-07-16",
      "results": {
        "applicable": 7,
        "passed": 7,
        "failed": 0,
        "skipped": 32,
        "scenarios_run": [
          "capability_discovery",
          "discovery",
          "error_handling",
          "health_check",
          "schema_compliance",
          "signals_flow",
          "validation"
        ]
      },
      "non_applicable_scenarios": [
        {
          "scenario": "media-buy / creative / governance / brand-rights tracks",
          "reason": "Require get_products, create_media_buy, sync_creatives, build_creative, si_*, check_governance, check_brand_rights tools that a signals-only agent does not expose. 32 scenarios skip cleanly per the runner's tool-gated applicability logic.",
          "upstream_issue": "https://github.com/adcontextprotocol/adcp/issues/2916",
          "upstream_status": "resolved — 5.25 runner emits capability-aware skip_if; signals-only agents no longer lose the error/schema/validation tracks."
        }
      ],
      "report": "docs/SEC42_ADCP_30_GA_COMPLIANCE.md",
      "reproduce": "API_KEY=$DEMO_API_KEY AGENT_URL=https://adcp.signal-stack.io/mcp npm run compliance"
    },
    "ucp": {
      "supported_spaces": [
        "openai-te3-small-d512-v1"
      ],
      "supported_encodings": [
        "float32"
      ],
      "dimensions": [
        512
      ],
      "embedding_endpoint_template": "/signals/{signal_id}/embedding",
      "similarity_search": true,
      "phase": "llm-v1",
      "gts_version": "adcp-gts-v1.0"
    },
    "dts": {
      "supported": true,
      "version": "1.2",
      "iab_techlab_compliant": true,
      "label_field": "x_dts",
      "privacy_compliance_mechanisms": [
        "TCF (Europe)",
        "GPP",
        "MSPA",
        "USPrivacy",
        "GPC"
      ],
      "supported_precision_levels": [
        "Individual",
        "Household",
        "Device",
        "Browser",
        "Geography"
      ],
      "offline_sources_supported": true,
      "provider_privacy_policy_url": "https://adcp-signals-adaptor.evgeny-193.workers.dev/privacy"
    },
    "id_resolution": {
      "supported": true,
      "cookieless_ready": true,
      "id_types_supported": [
        "cookie_3p",
        "maid_ios",
        "maid_android",
        "ctv_device",
        "uid2",
        "ramp_id",
        "id5",
        "hashed_email",
        "ip_only"
      ],
      "graph_partners": [
        {
          "name": "UID 2.0",
          "stage": "sandbox",
          "endpoint": null
        },
        {
          "name": "ID5",
          "stage": "sandbox",
          "endpoint": null
        },
        {
          "name": "LiveRamp RampID",
          "stage": "roadmap",
          "endpoint": null
        },
        {
          "name": "Yahoo ConnectID",
          "stage": "roadmap",
          "endpoint": null
        }
      ],
      "resolution_method": "derived_from_dts",
      "resolution_surface": "/signals/{signal_agent_segment_id} x_dts.data_sources + audience_precision_levels"
    },
    "measurement": {
      "supported": true,
      "reach_forecasting": {
        "supported": true,
        "endpoint": "/signals/{id} (detail panel)",
        "methodology": "logistic saturation against declared CPM + budget"
      },
      "overlap_analysis": {
        "supported": true,
        "endpoint": "/signals/overlap",
        "methodology": "category_affinity × min/max Jaccard heuristic"
      },
      "lift_measurement": {
        "supported": "mock",
        "partners_roadmap": [
          "Nielsen",
          "IAS",
          "DV",
          "Kantar",
          "Circana"
        ]
      },
      "delivery_simulation": {
        "supported": "mock",
        "surface": "/signals/{id} detail panel (post-activation)"
      }
    },
    "governance": {
      "audit_log": {
        "supported": true,
        "endpoint": "/mcp/recent",
        "retention_days": 7,
        "export_format": "csv_download"
      },
      "sensitive_category_flagging": {
        "supported": true,
        "categories_flagged": [
          "health",
          "financial",
          "political"
        ],
        "surface": "x_governance.sensitive block on signal + UI pill in signal detail"
      },
      "opt_out_url": "https://adcp-signals-adaptor.evgeny-193.workers.dev/privacy#opt-out",
      "audience_bias_governance_schema": {
        "supported": true,
        "version": "adcp_3.0"
      },
      "data_hygiene": {
        "supported": true,
        "schedule": {
          "cron": "0 6 * * SUN",
          "cadence": "weekly",
          "timezone": "UTC"
        },
        "retention": {
          "dynamic_signals_days": 7,
          "activation_jobs_days": 30,
          "tool_call_log_days": 7,
          "oauth_state_minutes": 10
        },
        "preserved": [
          "seeded signals (canonical catalog)",
          "derived signals (shipped with code)",
          "taxonomy nodes",
          "source records"
        ],
        "manual_trigger": {
          "method": "POST",
          "endpoint": "/admin/purge",
          "auth": "DEMO_API_KEY bearer"
        }
      }
    },
    "analytics": {
      "supported": true,
      "endpoints": {
        "query_vector": {
          "method": "POST",
          "path": "/ucp/query-vector"
        },
        "semantic_arithmetic": {
          "method": "POST",
          "path": "/ucp/arithmetic"
        },
        "analogy": {
          "method": "POST",
          "path": "/ucp/analogy"
        },
        "neighborhood": {
          "method": "POST",
          "path": "/ucp/neighborhood"
        },
        "coverage_gaps": {
          "method": "GET",
          "path": "/analytics/coverage-gaps"
        },
        "lorenz": {
          "method": "GET",
          "path": "/analytics/lorenz"
        },
        "summary": {
          "method": "GET",
          "path": "/analytics/summary"
        },
        "knn_graph": {
          "method": "GET",
          "path": "/analytics/knn-graph"
        },
        "seasonality": {
          "method": "GET",
          "path": "/analytics/seasonality"
        },
        "best_for_window": {
          "method": "GET",
          "path": "/analytics/best-for"
        },
        "portfolio_optimize": {
          "method": "POST",
          "path": "/portfolio/optimize"
        },
        "portfolio_pareto": {
          "method": "GET",
          "path": "/portfolio/pareto"
        },
        "portfolio_info_overlap": {
          "method": "POST",
          "path": "/portfolio/info-overlap"
        },
        "portfolio_from_brief": {
          "method": "POST",
          "path": "/portfolio/from-brief"
        },
        "portfolio_what_if": {
          "method": "POST",
          "path": "/portfolio/what-if"
        },
        "portfolio_hit_target": {
          "method": "POST",
          "path": "/portfolio/hit-target"
        },
        "audience_compose": {
          "method": "POST",
          "path": "/audience/compose"
        },
        "audience_saturation": {
          "method": "POST",
          "path": "/audience/saturation"
        },
        "audience_affinity_audit": {
          "method": "POST",
          "path": "/audience/affinity-audit"
        },
        "audience_journey": {
          "method": "POST",
          "path": "/audience/journey"
        },
        "audience_privacy_check": {
          "method": "POST",
          "path": "/audience/privacy-check"
        },
        "audience_holdout": {
          "method": "POST",
          "path": "/audience/holdout"
        },
        "audience_compose_ast": {
          "method": "POST",
          "path": "/audience/compose-ast"
        },
        "agents_directory": {
          "method": "GET",
          "path": "/agents/directory"
        },
        "agents_probe_all": {
          "method": "GET",
          "path": "/agents/probe-all"
        },
        "agents_orchestrate": {
          "method": "POST",
          "path": "/agents/orchestrate"
        },
        "agents_capability_matrix": {
          "method": "GET",
          "path": "/agents/capability-matrix"
        },
        "snapshot_save": {
          "method": "POST",
          "path": "/snapshots"
        },
        "snapshot_list": {
          "method": "GET",
          "path": "/snapshots"
        },
        "snapshot_get": {
          "method": "GET",
          "path": "/snapshots/:id"
        },
        "snapshot_delete": {
          "method": "DELETE",
          "path": "/snapshots/:id"
        },
        "snapshot_diff": {
          "method": "POST",
          "path": "/snapshots/diff"
        }
      },
      "methods": {
        "similarity_metric": "cosine",
        "arithmetic_model": "vector_l2_normalized",
        "analogy_algorithms": [
          "3cos_add",
          "3cos_mul"
        ],
        "portfolio_solver": "greedy_marginal_reach",
        "overlap_metrics": [
          "jaccard",
          "kl_divergence",
          "mutual_information"
        ],
        "projection_algorithms": [
          "jl_random",
          "umap_local_refine"
        ],
        "set_op_overlap_model": "category_affinity_jaccard_pairwise_inclusion_exclusion",
        "saturation_model": "poisson_1_minus_exp_neg_frequency",
        "affinity_index_basis": "reach_weighted_share_vs_catalog_baseline",
        "journey_model": "per_stage_compose_then_monotone_funnel",
        "privacy_model": "k_anonymity_floor_plus_sensitive_category_keywords",
        "holdout_model": "two_proportion_z_test_balanced_arms"
      },
      "limits": {
        "max_arithmetic_terms": 6,
        "max_portfolio_signals": 20,
        "max_neighborhood_k": 50,
        "max_query_vector_dim": 512,
        "min_signals_for_info_overlap": 2,
        "max_compose_signals": 20,
        "max_saturation_signals": 15,
        "max_affinity_audit_signals": 20
      },
      "signal_facets": {
        "x_analytics_fields": [
          "seasonality.monthly[12]",
          "seasonality.peakMonth",
          "decayHalfLifeDays",
          "volatilityIndex",
          "authorityScore",
          "idStabilityClass"
        ],
        "derivation": "deterministic_from_signal_metadata",
        "note": "Facets computed at mapper time; no DB migration required."
      }
    },
    "federation": {
      "supported": true,
      "protocol": "adcp_a2a_v0.1",
      "endpoints": {
        "registry": {
          "method": "GET",
          "path": "/agents/registry"
        },
        "federated_search": {
          "method": "POST",
          "path": "/agents/federated-search"
        },
        "cross_similarity": {
          "method": "POST",
          "path": "/agents/cross-similarity"
        },
        "reverse_taxonomy": {
          "method": "POST",
          "path": "/taxonomy/reverse"
        }
      },
      "partners": [
        {
          "id": "dstillery",
          "url": "https://adcp-signals-agent.dstillery.com/mcp",
          "stage": "live",
          "specialties": [
            "behavioral_audiences",
            "ttd_deployment"
          ]
        },
        {
          "id": "peer39",
          "stage": "roadmap",
          "specialties": [
            "contextual",
            "brand_safety"
          ]
        },
        {
          "id": "scope3",
          "stage": "roadmap",
          "specialties": [
            "sustainability"
          ]
        },
        {
          "id": "nextdata",
          "stage": "roadmap",
          "specialties": [
            "b2b_intent"
          ]
        }
      ],
      "alignment_method": "procrustes_svd_shared_anchors",
      "cache_ttl_seconds": 600
    }
  }
}

3 get_signals — the auth boundary key-gated

Sent without a key on purpose. Discovery and capabilities are open; the moment you ask for signal data, governance kicks in. The request shape:

{
  "jsonrpc": "2.0", "id": 3, "method": "tools/call",
  "params": {
    "name": "get_signals",
    "arguments": {
      "signal_spec": "in-market for premium CTV, US households, privacy-safe",
      "max_results": 2
    }
  }
}

And the honest answer without authorization:

{
  "jsonrpc": "2.0",
  "id": 3,
  "error": {
    "code": -32001,
    "message": "Authentication required for this method. Supply Authorization: Bearer <key>."
  }
}

4 get_products — the seller side (Cora AI Sales Agent) public

The same protocol, different vocabulary: a third-party sales agent with open discovery, asked for products against a brief ("premium CTV inventory, US", two results). This is what a buying agent receives to construct a media buy from — products, publishers, channels, format IDs with durations, delivery detail.

{
  "status": "completed",
  "message": "Found 2 CORA AI products.",
  "buying_mode": "brief",
  "products": [
    {
      "product_id": "prod-arcastation-ssp",
      "name": "Arcastation Korean FAST CTV",
      "description": "Guaranteed CTV inventory across Korean FAST channels with premium lean-back viewing.",
      "publisher": {
        "name": "Arcastation",
        "domain": "arcastation.com"
      },
      "domain": "arcastation.com",
      "channel": "ctv",
      "formats": [
        "video"
      ],
      "format_ids": [
        {
          "id": "video_16x9_30s",
          "source": "iab",
          "agent_url": "https://sales-agent.coraai.org/mcp/",
          "duration_ms": 30000
        },
        {
          "id": "ctv_video_30s",
          "source": "coraai",
          "agent_url": "https://sales-agent.coraai.org/mcp/",
          "duration_ms": 30000
        }
      ],
      "delivery_type": "guaranteed",
      "countries": [
        "KR",
        "US"
      ],
      "inventory_ids": [
        "inv-arcastation-ctv",
        "arcastation-ssp"
      ],
      "publisher_properties": [
        {
          "publisher_domain": "arcastation.com",
          "selection_type": "by_id",
          "property_ids": [
            "arcastation_ssp"
          ]
        }
      ],
      "pricing_options": [
        {
          "pricing_option_id": "po-arcastation-cpm-usd",
          "pricing_model": "cpm",
          "currency": "USD",
          "fixed_price": 15,
          "billing_event": "impression"
        }
      ],
      "pricing": [
        {
          "type": "cpm",
          "currency": "USD",
          "cpm": 15
        }
      ],
      "forecast": {
        "points": [
          {
            "label": "30 day availability",
            "metrics": {
              "impressions": {
                "mid": 6500000
              },
              "spend": {
                "mid": 97500
              }
            }
          }
        ],
        "forecast_range_unit": "availability",
        "method": "estimate",
        "currency": "USD",
        "measurement_source": "seller_ad_server",
        "generated_at": "2026-06-24T00:00:00Z"
      },
      "delivery_forecast": {
        "guaranteed_impressions": 6500000,
        "measurement_source": "seller_ad_server",
        "forecast_range_unit": "impressions",
        "confidence": "medium"
      },
      "reporting_capabilities": {
        "available_reporting_frequencies": [
          "daily"
        ],
        "expected_delay_minutes": 1440,
        "timezone": "UTC",
        "supports_webhooks": false,
        "available_metrics": [
          "impressions",
          "spend",
          "reach",
          "completed_views",
          "completion_rate",
          "viewability"
        ],
        "date_range_support": "date_range",
        "supports_geo_breakdown": {
          "country": true,
          "region": false
        },
        "supports_device_type_breakdown": true,
        "supports_placement_breakdown": true,
        "vendor_metrics": [
          {
            "vendor": {
              "domain": "scope3.com"
            },
            "metric_id": "emissions_gco2pm"
          }
        ]
      },
      "measurement_terms": {
        "billing_measurement": {
          "vendor": {
            "domain": "coraai.org"
          },
          "max_variance_percent": 10,
          "measurement_window": "standard"
        },
        "makegood_policy": {
          "available_remedies": [
            "additional_delivery",
            "credit"
          ]
        }
      },
      "targeting": {
        "devices": [
          "ctv"
        ],
        "environments": [
          "app"
        ],
        "languages": [
          "ko",
          "en"
        ]
      },
      "channels": [
        "ctv"
      ],
      "format_options": [
        {
          "format_option_id": "fo-prod-arcastation-ssp-video-30s",
          "display_name": "30 second hosted video",
          "format_kind": "video_hosted",
          "seller_preference": "preferred",
          "params": {
            "orientation": "horizontal",
            "aspect_ratio": "16:9",
            "min_width": 1280,
            "min_height": 720,
            "duration_ms_exact": 30000,
            "video_codecs": [
              "h264"
            ],
            "audio_codecs": [
              "aac"
            ],
            "containers": [
              "mp4"
            ],
            "max_file_size_mb": 500,
            "captions": "not_required",
            "asset_source": "buyer_uploaded",
            "buyer_asset_acceptance": "accepted"
          }
        }
      ],
      "brief_relevance": "Matched against buyer brief: premium CTV inventory, US"
    },
    {
      "product_id": "prod-kbs-fast",
      "name": "KBS FAST Channel CTV",
      "description": "Korean broadcast FAST supply for awareness campaigns targeting Korean-speaking audiences.",
      "publisher": {
        "name": "KBS",
        "domain": "kbs.co.kr"
      },
      "domain": "kbs.co.kr",
      "channel": "ctv",
      "formats": [
        "video"
      ],
      "format_ids": [
        {
          "id": "video_16x9_30s",
          "source": "iab",
          "agent_url": "https://sales-agent.coraai.org/mcp/",
          "duration_ms": 30000
        },
        {
          "id": "ctv_video_15s",
          "source": "coraai",
          "agent_url": "https://sales-agent.coraai.org/mcp/",
          "duration_ms": 15000
        }
      ],
      "delivery_type": "guaranteed",
      "countries": [
        "KR",
        "US"
      ],
      "inventory_ids": [
        "inv-kbs-fast-ctv",
        "kbs-fast"
      ],
      "publisher_properties": [
        {
          "publisher_domain": "kbs.co.kr",
          "selection_type": "by_id",
          "property_ids": [
            "kbs_fast"
          ]
        }
      ],
      "pricing_options": [
        {
          "pricing_option_id": "po-kbs-cpm-usd",
          "pricing_model": "cpm",
          "currency": "USD",
          "fixed_price": 18,
          "billing_event": "impression"
        }
      ],
      "pricing": [
        {
          "type": "cpm",
          "currency": "USD",
          "cpm": 18
        }
      ],
      "forecast": {
        "points": [
          {
            "label": "30 day availability",
            "metrics": {
              "impressions": {
                "mid": 4200000
              },
              "spend": {
                "mid": 75600
              }
            }
          }
        ],
        "forecast_range_unit": "availability",
        "method": "estimate",
        "currency": "USD",
        "measurement_source": "seller_ad_server",
        "generated_at": "2026-06-24T00:00:00Z"
      },
      "delivery_forecast": {
        "guaranteed_impressions": 4200000,
        "measurement_source": "seller_ad_server",
        "forecast_range_unit": "impressions",
        "confidence": "medium"
      },
      "reporting_capabilities": {
        "available_reporting_frequencies": [
          "daily"
        ],
        "expected_delay_minutes": 1440,
        "timezone": "UTC",
        "supports_webhooks": false,
        "available_metrics": [
          "impressions",
          "spend",
          "reach",
          "completed_views",
          "completion_rate",
          "viewability"
        ],
        "date_range_support": "date_range",
        "supports_geo_breakdown": {
          "country": true,
          "region": false
        },
        "supports_device_type_breakdown": true,
        "supports_placement_breakdown": true,
        "vendor_metrics": [
          {
            "vendor": {
              "domain": "scope3.com"
            },
            "metric_id": "emissions_gco2pm"
          }
        ]
      },
      "measurement_terms": {
        "billing_measurement": {
          "vendor": {
            "domain": "coraai.org"
          },
          "max_variance_percent": 10,
          "measurement_window": "standard"
        },
        "makegood_policy": {
          "available_remedies": [
            "additional_delivery",
            "credit"
          ]
        }
      },
      "targeting": {
        "devices": [
          "ctv"
        ],
        "environments": [
          "app"
        ],
        "languages": [
          "ko"
        ]
      },
      "channels": [
        "ctv"
      ],
      "format_options": [
        {
          "format_option_id": "fo-prod-kbs-fast-video-30s",
          "display_name": "30 second hosted video",
          "format_kind": "video_hosted",
          "seller_preference": "preferred",
          "params": {
            "orientation": "horizontal",
            "aspect_ratio": "16:9",
            "min_width": 1280,
            "min_height": 720,
            "duration_ms_exact": 30000,
            "video_codecs": [
              "h264"
            ],
            "audio_codecs": [
              "aac"
            ],
            "containers": [
              "mp4"
            ],
            "max_file_size_mb": 500,
            "captions": "not_required",
            "asset_source": "buyer_uploaded",
            "buyer_asset_acceptance": "accepted"
          }
        }
      ],
      "brief_relevance": "Matched against buyer brief: premium CTV inventory, US"
    }
  ],
  "count": 2,
  "total": 22,
  "currency": "USD",
  "cache_scope": "public",
  "wholesale_feed_version": "coraai-products-2026-06-24",
  "pricing_version": "coraai-usd-pricing-2026-06-24",
  "pagination": {
    "has_more": true
  },
  "property_list_applied": false,
  "brief": "premium CTV inventory, US",
  "filter_diagnostics": {
    "total_candidates": 4,
    "matched_candidates": 22,
    "no_match_targeting": false,
    "semantics": "approximate"
  },
  "adcp_version": "3.1",
  "adcp_major_version": 3
}

The protocol vocabulary, as observed

Not the spec's aspiration — the tool names actually exposed by the 7 open-discovery agents, aggregated from the 2026-07-22 sweep and grouped by agent type. ×N marks a name seen at N different agents: the ×2s are where the shared protocol surface is already forming — get_products, create_media_buy, sync_creatives, list_creative_formats on the transaction side, get_signals and get_adcp_capabilities everywhere. Full context in the AdCP guide.

Sales 4 open agents · 37 distinct tools

get_adcp_capabilities ×2get_products ×2list_creative_formats ×2list_creatives ×2activate_signalcalibrate_contentcomply_test_controllercreate_collection_listcreate_content_standardscreate_media_buycreate_property_listdelete_content_standardsdelete_property_listexplain_rulefix_vastget_content_standardsget_media_buy_deliveryget_media_buysget_property_listget_signalslist_accountslist_collection_listslist_content_standardslist_property_listslist_rulespreview_creativesync_accountssync_creativessync_governanceupdate_content_standardsupdate_media_buyupdate_property_listvalidate_content_deliveryvalidate_inputvalidate_property_deliveryvalidate_vastvalidate_vast_url

Signals 2 open agents · 8 distinct tools

get_adcp_capabilities ×2get_signals ×2activate_signalget_conceptget_operation_statusget_similar_signalsquery_signals_nlsearch_concepts

Buying 1 open agent · 11 distinct tools

create_media_buyget_media_buy_deliveryget_media_buysget_productslist_authorized_propertieslist_creative_formatslist_creativessync_creativestasks_gettasks_listupdate_media_buy

adagents.json — who is allowed to sell

The registry answers "which agents exist." The AdCP spec's /.well-known/adagents.json answers the other half: a domain declares which agents are authorized to represent it — the agentic-era analogue of ads.txt, so a buying agent can verify the counterparty before transacting. Not a mock: this is the manifest served live by this site's own agent domain (captured 2026-07-22):

{
  "$schema": "https://adcontextprotocol.org/schemas/v3/adagents.json",
  "contact": {
    "name": "AdCP Signals Adaptor — Demo Provider (Evgeny)",
    "email": "Evgeny@gmail.com",
    "domain": "adcp.signal-stack.io",
    "privacy_policy_url": "https://adcp.signal-stack.io/privacy"
  },
  "authorized_agents": [
    {
      "url": "https://adcp.signal-stack.io/mcp",
      "authorized_for": "AdCP signals data provider for the Demo Provider catalog (audience + outcome signals across 14 verticals)",
      "authorization_type": "signal_tags",
      "signal_tags": [
        "all"
      ]
    }
  ],
  "signal_tags": {
    "all": {
      "name": "All catalog signals",
      "description": "Every signal in the Demo Provider catalog, spanning Demographic / Interest / Purchase intent / Geo / Composite category types across 14 verticals. Includes both deterministic (rule-based) and dynamic (proposal-derived) signals."
    }
  },
  "last_updated": "2026-07-22T16:22:29.525Z"
}

For a publisher declaring third-party sales agents, the spec's shape looks like this (illustrative, per the AdCP standards guide):

{
  "publisher": "example-publisher.com",
  "authorized_agents": [
    {
      "url": "https://sales-agent.example-ssp.com",
      "role": "seller",
      "properties": ["news", "sports"]
    }
  ],
  "signing_keys": [
    { "kid": "2026-key-1", "alg": "Ed25519" }
  ]
}

Check any publisher domain

Opens https://<domain>/.well-known/adagents.json in a new tab. A 404 is a finding too: the domain hasn't declared its authorized agents yet. The sibling manifest for brands — brand.json — is live on this site.

Spec & conformance state

Latest spec releasev3.1.8published 2026-07-28

Latest prereleasev3.1.0-rc.15release-candidate line

Conformance SDK · 3.07.11.7stable line

Conformance SDK · 3.112.1.1current line

7/7 compliance scenarios passing This site's own registered signals agent, run against the conformance suite on every watcher cycle: capability_discovery · discovery · error_handling · health_check · schema_compliance · signals_flow · validation. Per-agent conformance isn't published by the registry — the honest way to check any agent is exchange 2 above: call its capabilities and read supported_versions for yourself.

Count over time

2026-07-2221 agents

The series starts at first observation and appends on every change — a dated, append-only record of the ecosystem actually growing, kept honest the same way the Living Maps are.

Who has actuallyshipped an agent.