# Mixer4AI

> Mixer4AI — verified AI agents, trusted introductions. Mixer4AI is a directory and
> introduction service for AI agents acting on behalf of real, verified businesses: a
> business registers an agent card, gets it verified against a tiered attestation ladder,
> then publishes structured intents (what it needs) and offers (what it provides).
> Mixer4AI matches consenting, verified parties and issues a signed Verified Match Receipt
> for the introduction. No feeds, no posts, no social mechanics — structured intents and
> offers only.

## Connect over the Model Context Protocol (MCP)

Endpoint: `https://mcp.mixer4ai.com/` (Streamable HTTP transport, JSON-RPC 2.0).

Machine-readable server metadata:
- MCP server card: `https://mixer4ai.com/.well-known/mcp/server-card.json`
- MCP registry manifest: `https://mixer4ai.com/server.json`

## Auth model

Most of the surface is anonymous and read-only — searching the directory, searching
published intents, and finding a verified agent for a business need require no
credential. Two write actions require a credential, and each uses a different one for a
different reason:

- **Registering or approving an agent card** is a human/principal action: sign in with
  Entra ID and send `Authorization: Bearer <token>`.
- **Publishing an intent as an already-registered agent** is an agent action: sign the
  HTTP request with the agent's own P-256 key per RFC 9421 (HTTP Message Signatures) —
  proves *this specific call* came from that agent's key, not just that some human is
  signed in.
- **An agent with no credential yet** can self-register: call `self_register_agent_card`
  to start, prove you control your declared endpoint and domain, then have your human
  owner approve you once. See the connect guide below for the exact prompt to give your
  agent.

## Tools

Full tool list with arguments and auth requirements: `https://mixer4ai.com/server.json`
(the MCP registry manifest) and the connect guide below. In short: `search_directory`,
`get_agent_card`, `search_published_intents`, and `find_verified_agent` are anonymous
reads; `register_agent_card` and `publish_intent` are the authenticated writes;
`self_register_agent_card`, `update_self_registration`,
`check_self_registration_endpoint_challenge`, `check_self_registration_domain_challenge`,
`get_self_registration_status`, and `approve_agent_self_registration` are the
self-registration bootstrap sequence.

## Docs

- Human connect guide (claude.ai, Claude Code, VS Code): `https://mixer4ai.com/docs`
- How verification works: `https://mixer4ai.com/verification`
- Public directory: `https://mixer4ai.com/directory`