MCP server

AdvocateMCP

A Model Context Protocol server that lets any MCP-enabled AI assistant search a directory of AI-ready local businesses and query each one’s canonical agent in real time. Neutral, platform-agnostic, zero setup per AI.

Endpoint

URLhttps://api.advocatemcp.com/mcp

TransportStreamable HTTP (JSON-RPC 2.0 over HTTP POST, SSE for server-initiated messages)

AuthenticationNone, for every tool. Reservations, callbacks, subscriptions and handoffs are all reachable without a credential. Abuse is bounded by idempotency keys, short auto-expiring holds and per-IP rate limits rather than by authentication.

What it does

Local businesses (plumbers, law firms, agencies, copywriters, etc.) subscribe to Advocate and fill in a structured profile: services, pricing, hours, credentials, customer quotes, lead-routing preferences. Advocate publishes each profile as a standards-compliant AI advocate agent.

When a user asks their AI assistant “find me a good plumber in Boise who does 24/7 emergency,” the assistant calls search_businesses, then query_business_agent on the top results. The agent answers in the business’s own voice, with verified metadata and a trackable referral link. The AI assistant sees citation-ready answers instead of scraping a dozen inconsistent review sites.

For richer interactions, the read-only planning tools (get_availability, get_quote) and the transactional tools (reserve_slot, request_callback, subscribe_to_updates, initiate_handoff) let the AI assistant act on behalf of the user: check an opening-hours window, get a price range, hold a slot, or hand the conversation off to a human.

Tools

Tool	Description
search_businesses	Search for registered businesses by category, name, or location. Returns slugs + agent endpoints.
query_business_agent	Query a specific business’s AI advocate agent. Returns a citation-ready answer with referral link.
get_availability	30-minute windows derived from the business’s configured operating hours — not a live booking calendar. A returned window means the business is normally open then, not that the slot is confirmed free. Real-calendar integration is on the roadmap.
get_quote	Service price quote from `pricing_json_v2`. Returns an exact, range, or estimate label.
reserve_slot	Create a 15-minute HELD reservation; returns a signed `confirmation_token` the agent posts back to `/a2a/confirm`.
initiate_handoff	Start a handoff to a human (SMS/email via tenant routing) or another agent (signed continuation URL).
get_credentials	The business’s self-reported licenses, insurance, bonding and certifications. Framed explicitly as self-reported so agents don’t restate tenant claims as verified fact.
get_cancellation_policy	The business’s cancellation / refund / no-show policy, verbatim, so an agent can quote it. Returns guidance when the business hasn’t set one.
request_callback	Push a user’s contact details to the business. Idempotent on `(slug, idempotency_key)` within 24h, so agent retries don’t spam the business.
subscribe_to_updates	Subscribe a user’s email to updates from a business. Double opt-in (CAN-SPAM / GDPR): returns a confirmation link, never auto-activates, token expires in 7 days.

The full machine-readable capability manifest, including input schemas, rate-limit declarations, and the list of business slugs, is published at /.well-known/mcp.json.

Quick start

Claude Desktop

Add this block to your claude_desktop_config.json:

{
  "mcpServers": {
    "advocate": {
      "url": "https://api.advocatemcp.com/mcp"
    }
  }
}

Cursor

Settings → MCP Servers → Add server, then paste the URL above.

Any MCP client

Point at https://api.advocatemcp.com/mcp. The server advertises its tools via the standard initialize → tools/list handshake; no registration is required for discovery or search.

Rate limits

MCP tools (/mcp): 60 requests per minute per IP, applied flat across all ten tools. There is no separate, tighter limit for the transactional tools.
A2A endpoints (/a2a/*) are served by a different origin, which keys the limit one of two ways — not both: per-agent when an x-agent-identity header is present (100/min unverified, 250/min known, 1000/min trusted), otherwise per-IP at 100/min. Neither applies to /mcp.
Rate limiting is enforced at the edge by a Cloudflare Durable Object; exceeding it returns 429 with a Retry-After header.

Attribution

Every referral URL returned by query_business_agent is a signed tracking token of the form https://api.advocatemcp.com/r/<token>. Clicks are recorded with the originating AI assistant (via x-agent-identity) and the tenant can see per-AI conversion data in their dashboard. No cross-site cookies; IP addresses are salted-hashed and discarded within 30 days.

For businesses

If you run a local business and want to be findable through this server, run a free citation audit to see where AI assistants currently land when asked about your category, or head to advocatemcp.com for the 15-minute onboarding.

Status. This server is live and handling production traffic. It is listed on the official MCP Registry (com.advocatemcp/advocate) and on Smithery. Questions or integration help: support@advocatemcp.com.