Skip to main content
Version: 3.1.0

Automotive, ready for agents

Auto Agent Protocol gives AI agents a shared language for automotive retail. Discover a dealer, search actual inventory, inspect evidence and pricing, then send a lead with the buyer's permission.

MCP, A2A and PAP connect buyer agents to one shared automotive contractOpen full-size diagram

AAP 3.1 defines three native interfaces. Each carries the same typed request and response; none requires an extra hop through another protocol.

InterfaceHow the agent callsBaseline
MCPDiscover and call automotive toolsStreamable HTTP, 2026-07-28 + 2025-11-25 compatibility
A2ASendMessage with an automotive DataPartA2A 1.0, JSON-RPC 2.0
PAPNative OpenAPI operation with Session + DPoPPersonal Agent Protocol draft 0.3

A full AAP 3.1 dealer deployment implements all three with the same skill subset and business behavior. A client needs only its chosen interface. A partial implementation can describe the binding it supports, but cannot claim the full deployment profile. Conformance defines the exact boundary.

Five skills. One shopping journey.​

The five automotive skills move from dealer discovery and inventory to evidence-backed detail and a consented leadOpen full-size diagram

SkillAnswer
dealer.informationWho is this dealer, and where are its locations?
inventory.facetsWhich filters and feature codes are useful here?
inventory.searchWhich vehicles meet every supported constraint?
inventory.vehicleWhat is known about this specific listing?
lead.submitCan the dealer receive this explicitly authorized inquiry?

Declare one or more implemented skills. Search with feature filters also requires facets for its feature vocabulary. For a complete shopping experience, implement all five. The same vehicle model covers cars, EVs, motorcycles, RVs and trailers; an omitted vehicle_type means car.

Automotive meaning, shared everywhere​

The AAP automotive layer sits above three independent protocol bindings and one dealer backendOpen full-size diagram

  • Search: explicit capabilities, verified features, radius/distance semantics and complete cursor pagination.
  • Truthful listings: availability, freshness, identifiers and evidence; missing data remains unknown.
  • Clear pricing: one authoritative advertised price; mandatory dealer charges are included, only government-required consumer charges are separate.
  • Explicit units: native USD/CAD prices and mi/km odometers; single-currency searches and exact odometer comparisons. Omitted labels preserve USD/mi.
  • Controlled contact: explicit channel consent, mandatory retry keys and truthful appointment outcomes.

AAP covers discovery, research and dealer contact. Payments, financing approval, transaction quotes, reservations and trade-in valuations are outside this release.

Start on the dealer's domain​

Publish dealer-domain discovery, connect the three interfaces, verify the shared behavior and expose the working serviceOpen full-size diagram

Publish /.well-known/agent-card.json, /.well-known/poppy.json and /llms.txt on the dealer's HTTPS website. They link the actual MCP endpoint, A2A interface and deployment-specific PAP OpenAPI. Services can run on a provider's tenant infrastructure.

An agent follows an explicit endpoint and its access requirements. It does not have to guess a URL or scrape a search form. Discovery improves access; it does not guarantee every assistant will automatically use a protocol.

The selected automotive contract is identified by:

https://autoagentprotocol.org/extensions/aap/v3.1

For a quick implementation path, read discovery, choose the MCP, A2A or PAP client guide, and validate against the schemas. Existing operators should start with the v3 migration checklist.