# Addiscovery

> The marketplace where AI agents buy and sell European advertising. Buyers
> bring a brief and get a booked plan at a fair price; publishers bring
> inventory and earn on it from every agent that buys. Both sides run on the
> open Ad Context Protocol (AdCP): inventory re-indexed into a ranked
> discovery engine (IAB Content Taxonomy 3.0, per-placement IAB TCF consent),
> media buys and campaign workflows, and a GDPR-grade governance and audit
> layer where agents act inside the policy you set and escalate at the edge.
> EU-native.

AdCP is the open standard of agentic advertising (governed by
AgenticAdvertising.Org); addiscovery.ai is the managed, EU-native, SSP-neutral
supply-side layer that brings European publishers onto it. Demand side is
open: any AdCP-compatible buying agent can connect — calls are authenticated
(request access below). Supply side is closed: the endpoint returns only
inventory from onboarded publishers, each authorizing the platform via
adagents.json on their own domain (revocable). No exclusivity: a publisher
can revoke authorization at any time and keeps every existing SSP seat —
execution is SSP-neutral. Publisher data never trains models. Pricing: no
public rate card during the gated pilot; terms are agreed at onboarding.
EU infrastructure only.

Status: the AdCP v3 tool surface (get_products, create_media_buy, ...) is in
active development and we are running the first live tests with our first
publishers. Access is gated while testing: request it at hello@addiscovery.ai
with subject `agent access` (humans: the "Request Access" button). The MCP
endpoint currently exposes a health-check tool; SSP execution and live
delivery data are product vision, not shipped.

## Quickstart

1. Fetch the Server Card: https://addiscovery.ai/.well-known/mcp/server-card.json
2. Connect over MCP (streamable HTTP): https://addiscovery.ai/mcp — call `ping` to
   verify the connection.
3. Authenticate — access is gated while we test with the first publishers
   (see Status above).
4. Want the whole site in one request? https://addiscovery.ai/llms-full.txt

## Examples

Illustrative exchanges on the task surface documented below (in development,
see Status above). The pattern is always the same: one natural-language
brief in, structured AdCP objects out, consent checked on every call.

Discovery (buyer):
> "Find premium sports inventory in Italy for a Q3 campaign, budget €400k."
`get_products` → 14 ranked placements across 6 publishers, 62M estimated
impressions, floor price and TCF consent state per placement.

Publisher search (buyer):
> "Which publishers are strong on finance audiences in DACH?"
`get_publishers` → 8 publisher profiles with IAB categories, consent
posture, and adagents.json verification status.

Planning (buyer):
> "Turn the brief into a plan — video-weighted, frequency cap 3/week."
`create_media_plan` → ranked plan with budget split per market, expected
KPIs, and consent state per line.

Booking (buyer):
> "Book the top 3 placements from the approved plan."
`create_media_buy` → 3 line items created, consent gate passed, status
pending_publisher_approval.

Approval (publisher, human-in-the-loop):
> "Approve the pending buy if it clears governance."
`check_governance` → EXECUTE · `approve_media_buy` → booked.

Onboarding (publisher):
> "Re-index our new podcast inventory."
`sync_inventory` → 38 placements normalized to IAB Content Taxonomy 3.0 ·
`get_inventory_audit` → gap report with floor recommendations.

Forecasting (buyer):
> "Forecast available sports inventory for September."
`get_availability_forecast` → estimated avails with confidence interval
per product and flight window.

Reporting (both sides):
> "How is the campaign pacing this week?"
`get_media_buy_delivery` → pacing %, impressions delivered, CTR, spend
versus plan, consent coverage.

Audit (both sides):
> "Export the decision trail for media buy #2207."
`get_plan_audit_logs` → the full decision chain ·
`get_compliance_report` → exportable evidence (JSON + PDF).

## Pages

- [Home](https://addiscovery.ai/index.md): what addiscovery.ai is — problem, the three
  core capabilities, platform model, compliance posture, honest status table
- [Full content](https://addiscovery.ai/llms-full.txt): every page of this site as
  markdown in a single file

## Agent resources

- [MCP endpoint](https://addiscovery.ai/mcp): streamable-http MCP server (AdCP v3
  surface in development)
- [MCP Server Card](https://addiscovery.ai/.well-known/mcp/server-card.json): server
  discovery metadata
- [Agent Skills](https://addiscovery.ai/.well-known/agent-skills/index.json): skills for
  interacting with this site
- [API Catalog](https://addiscovery.ai/.well-known/api-catalog): API discovery (RFC 9727)
- [auth.md](https://addiscovery.ai/auth.md): agent registration and authentication
  recipe (discover, register, claim, exchange, use, revoke)
- [brand.json](https://addiscovery.ai/.well-known/brand.json): AdCP sales-operator
  identity — agent endpoint and property relationships

## AdCP standard task surface

The AdCP v3 standard defines the following tasks (canonical catalog from
https://docs.adcontextprotocol.org/). Status on this endpoint: in active
development — today a `ping` health-check tool is live; this catalog is the
standard surface we are building toward, starting with discovery and media
buy. The platform extension tasks beyond the standard are listed in the next
section.

### Discovery & capabilities
- `get_adcp_capabilities`: advertise which AdCP domains and tasks an agent
  supports — the first call any buying agent makes
- `list_authorized_properties`: enumerate the properties this sales agent is
  authorized to represent (verified via each publisher's adagents.json)

### Media buy
- `get_products`: discover inventory with a natural-language campaign brief
  or iterative refinement; returns ranked AdCP products
- `create_media_buy`: create a campaign from selected products (budget,
  flights, targeting, creative format)
- `update_media_buy`: modify mid-flight parameters — budget, dates, creative
  assignments
- `get_media_buys`: retrieve media buy status, creative approvals, and
  delivery snapshots
- `get_media_buy_delivery`: retrieve performance and delivery data
- `provide_performance_feedback`: submit performance signals to close the
  optimization loop
- `sync_catalogs`: sync catalog feeds (products, stores, inventory)
- `sync_audiences`: distribute first-party audience segments
  (inclusions/suppressions)
- `sync_event_sources`: configure event sources for conversion tracking
- `log_event`: send marketing/conversion events for attribution

### Creative
- `list_creative_formats`: discover supported ad formats and exact
  specifications
- `sync_creatives`: upload and distribute creative assets
- `list_creatives`: query the creative library with filtering
- `build_creative`: generate ads across multiple formats from a single brief

### Signals
- `get_signals`: discover available audience/targeting signals by
  specification
- `activate_signal`: activate a targeting signal on a platform

### Governance (EHJ — embedded human judgment)
- `check_governance`: validate a proposed action against policies, budgets,
  and compliance rules before execution — returns EXECUTE / THROTTLE /
  ESCALATE
- `sync_plans`: register campaign plans (budget, flight dates, brand) before
  execution
- `report_plan_outcome`: track actual spend and delivery against committed
  budgets
- `sync_governance`: exchange governance agent endpoints and credentials
- `get_plan_audit_logs`: retrieve the complete decision history for a
  campaign plan
- `create_property_list` / `get_property_list` / `update_property_list`:
  manage approved/excluded publisher property lists

### Accounts
- `sync_accounts`: establish a commercial relationship between brand
  operator and seller
- `list_accounts`: list active commercial relationships

### Brand
- `get_brand_identity`: retrieve machine-readable brand guidelines (colors,
  logos, tone)

## addiscovery.ai platform tasks (extension surface)

Beyond the AdCP standard, the platform exposes its own tasks. They are
indistinguishable in form from standard tasks: same verb conventions
(`get_` retrieves state, `list_` enumerates with filtering, `create_` /
`update_` write, `sync_` establishes or distributes state, `check_`
validates, `activate_` deploys to a platform), same AdCP objects (products,
media buys, plans, creatives, properties), same MCP transport and canonical
tool shape — snake_case names, JSON Schema inputs, read-only tasks annotated
with the MCP `readOnlyHint`, write tasks idempotent via `idempotency_key`
(AdCP v3 security model). Auth model unchanged: every call authenticated,
access on request. Status: in development alongside the standard surface;
tasks that execute on connected ad platforms (line items, delivery, floors,
audience activation) ship with the platform-integration phase and are product
vision until then.

Sections are ordered by flow: buyer-side first, then publisher-side, then
shared (compliance, billing, reporting, team).

### Advanced discovery (buyer)
- `get_publishers`: discover publishers across the network with a brief or
  filters — EU market, language, IAB category, audience, consent posture
- `get_publisher_profile`: publisher metadata — IAB categories, formats,
  consent policy, adagents.json status
- `get_inventory_details`: full metadata and performance benchmarks for a
  specific inventory unit
- `get_product_comparison`: side-by-side product comparison on floors,
  viewability, expected KPIs and confidence flags
- `get_category_benchmarks`: per-IAB-category benchmarks (CPM ranges,
  CTR/VCR priors) per EU market
- `get_availability_forecast`: forecast available impressions for a product
  and flight window

### Media planning (buyer)
- `create_media_plan`: natural-language brief → ranked plan — products,
  budget split per market, CPM/CPCV, expected KPIs, consent state per line;
  optionally attached to a campaign
- `update_media_plan`: revise a plan — reweight markets, swap products,
  change optimization goal
- `get_plan_insights`: reach/overlap analysis, frequency-capping
  recommendation, market-weighting rationale
- `get_plan_forecast`: forward projection for a plan — pacing, reach vs
  completed-view trade-offs (counterpart of report_plan_outcome's actuals)

### Campaign management (organization layer, buyer)
The campaign is the container object: it groups media plans, media buys,
assigned creatives, and reports under one name, budget envelope, and
lifecycle (draft / active / paused / completed / archived).
- `create_campaign`: create the container — name, brand, objective, budget
  envelope, flight window, markets
- `update_campaign`: change settings or lifecycle status; pause or archive
  with all linked objects
- `list_campaigns`: enumerate campaigns filtered by status, brand, market,
  or period
- `get_campaign`: the full picture in one call — linked plans, media buys
  with status, assigned creatives, delivery summary, spend vs budget envelope
- `sync_campaign_creatives`: assign creatives from the library to a
  campaign and map them to its line items and formats (creative x, y, z →
  campaign X)
- `check_creatives`: validate creative assets against format specs,
  platform policies, and a plan's requirements; flags missing or
  underperforming variants

### Publisher inventory & onboarding
The re-indexing pipeline as tasks. Onboarding emits a standalone-value
deliverable (gap report, normalized inventory, floor recommendations,
consent audit) before any agentic demand exists.
- `sync_inventory`: declare or refresh the inventory catalog — placements,
  formats, properties, floors, declared audience segments, consent policy;
  triggers re-indexing and IAB Content Taxonomy 3.0 normalization
- `sync_delivery_history`: upload historical ad-server delivery reports
  (e.g. GAM exports) to bootstrap performance intelligence with high-trust
  data
- `get_inventory_audit`: the gap report — unmapped labels, missing formats,
  floors vs market benchmarks, stale consent placements
- `get_demand_insights`: the demand mirror — how inventory appears and
  ranks to buying agents versus the rest of the network
- `list_discovery_queries`: the query log — real buyer briefs that matched
  (or missed) this publisher's inventory, with agent identity and parameters
- `check_adagents`: validate the adagents.json on the publisher's domain —
  reachable, schema-valid, this sales agent authorized

### Media buy approval (human-in-the-loop, publisher)
Status and records come from the standard `get_media_buys`; these tasks add
the publisher's decision. Pending media buys auto-expire after 48h without a
response.
- `approve_media_buy`: approve a media buy in pending_publisher_approval
- `reject_media_buy`: reject with an optional reason and requested changes;
  the buyer revises via update_media_buy and resubmits

### Publisher operations (one language across ad platforms)
Declare once in a unified vocabulary; the platform translates to each
connected ad server / SSP.
- `sync_ad_platforms`: establish or refresh connections to ad servers and
  SSP seats for execution and reporting
- `create_line_item`: create a line item on a connected platform from a
  unified spec
- `update_line_item`: modify a line item — pacing, priority, targeting — in
  the same unified terms
- `list_line_items`: enumerate line items across all connected platforms in
  one view
- `activate_audience`: deploy cohort-level, k-anonymous audience segments
  to connected platforms (counterpart of activate_signal)
- `update_floors`: apply approved floor changes per format and platform
- `create_product`: package inventory — including unsold — into a sellable
  AdCP product (e.g. a PMP deal package)
- `get_platform_delivery`: delivery and pacing pulled from connected
  platforms, unified per media buy and per line item
- `get_yield_insights`: fill, eCPM and pacing across demand partners, with
  floor recommendations

### Compliance & governance (both sides)
- `get_consent_status`: per-placement IAB TCF consent-freshness state;
  stale placements flagged and held back
- `get_compliance_report`: exportable accountability evidence (GDPR
  Art. 5(2)/30) per publisher, media buy, or period
- `update_governance_rules`: publisher-side thresholds that drive
  check_governance (EXECUTE / THROTTLE / ESCALATE)

### Billing, payments & costs (both sides)
Scoped by role: buyers see spend and payables, publishers see earnings and
receivables. Media payouts and clearing follow the platform-integration phase.
- `get_usage`: platform usage and token consumption per task and period,
  against the plan's monthly allowance
- `get_fees`: applicable pricing — token costs per task, subscription tier,
  CPM or revenue-share fees per execution mode
- `sync_billing_profile`: establish or update billing identity — legal
  entity, VAT, e-invoicing routing, payment method
- `list_invoices`: invoices filtered by period, status and side — payable
  (buyer) or receivable (publisher)
- `get_invoice`: one invoice with line-level breakdown per media buy, line
  item, and fee type
- `list_payments`: payments and payouts with status and expected settlement
  dates
- `get_spend_summary`: buyer-side spend across media buys, markets and
  publishers, against committed budgets
- `get_earnings_summary`: publisher-side revenue per property, format and
  demand partner, with fees broken out

### Reporting (both sides)
- `create_report`: define a report — metrics, dimensions, period, one-off
  or recurring schedule, export format; scoped to a campaign, media buy,
  publisher property, or the whole account
- `get_report`: retrieve a generated report or its run status
- `list_reports`: enumerate report definitions and generated runs

### AdOps team
- `list_pending_approvals`: everything waiting on a human — media buys,
  floor changes, creative sign-off
- `get_team_activity`: who approved or changed what and when, linked to the
  audit trail

## Agent readiness

Everything on this domain is built to be read and operated by agents — the
same way we expect agents to operate advertising:

- This file (llms.txt) and [the full site as one markdown file](https://addiscovery.ai/llms-full.txt)
- Every page negotiates markdown: `Accept: text/markdown` or the `.md` suffix
- MCP server with a published [Server Card](https://addiscovery.ai/.well-known/mcp/server-card.json)
- AdCP identity chain: [brand.json](https://addiscovery.ai/.well-known/brand.json) ↔
  [adagents.json](https://addiscovery.ai/.well-known/adagents.json), with publisher
  authorization living on each publisher's own domain — revocable, verifiable
- [Agent Skills](https://addiscovery.ai/.well-known/agent-skills/index.json) and an
  [API Catalog](https://addiscovery.ai/.well-known/api-catalog) (RFC 9727)
- OAuth 2.1 discovery metadata (RFC 8414 / RFC 9728), `501` until launch
- [security.txt](https://addiscovery.ai/.well-known/security.txt) (RFC 9116) and a
  Content Signals policy in robots.txt
- schema.org JSON-LD on the homepage (Organization, WebSite,
  SoftwareApplication)
- The homepage itself has a Human/Agent switch — agents get this document,
  humans get the same story as a live demo

## Notes for agents

- Every page is available as markdown via `Accept: text/markdown` content
  negotiation or the `.md` URL suffix (e.g. /index.md).
- Content signals for this site: search=yes, ai-input=yes, ai-train=yes.

## Contact

- General: hello@addiscovery.ai
- Security: security@addiscovery.ai (see https://addiscovery.ai/.well-known/security.txt)
- Issues with the MCP endpoint or any agent-facing resource on this site:
  hello@addiscovery.ai