buddhiAI
Browse the docs

Connecting via Claude

Add Buddhi AI as a connector in Claude.ai — no API key to copy.

If you're adding Buddhi AI from Claude.ai (as a custom connector or from the Connectors Directory), you don't need an agent key at all. Claude authenticates through OAuth and mints its own key for you automatically.

Adding the connector

  1. In Claude.ai, go to Settings → Connectors → Add custom connector (or, once listed, find Buddhi AI in the Connectors Directory and click Add).
  2. Enter the server URL: 
    https://your-buddhi-host/mcp
  3. Claude opens a browser window to Buddhi AI's sign-in and consent screen. Sign in (or create an account) and review what access is being granted — the screen shows exactly which tools the connection will have and which Buddhi AI user it will act as.
  4. Click Approve. Claude is redirected back and the connector is ready — no key to copy, store, or rotate yourself.

What happens behind the scenes

sequenceDiagram
    participant Claude as Claude.ai
    participant Auth as Buddhi AI (oauth/authorize)
    participant Gateway as MCP Gateway (/mcp)

    Claude->>Gateway: POST /mcp (no auth)
    Gateway-->>Claude: 401 + WWW-Authenticate (resource metadata)
    Claude->>Auth: Discover authorization server, register client
    Claude->>Auth: Open /oauth/authorize (PKCE)
    Auth-->>Claude: User signs in, reviews consent, approves
    Auth->>Claude: Redirect with authorization code
    Claude->>Auth: POST /oauth/token (code + PKCE verifier)
    Auth-->>Claude: Bearer access token (a normal agent key)
    Claude->>Gateway: POST /mcp (Authorization: Bearer ...)
    Gateway-->>Claude: Tool results, scoped to your role

This is standard OAuth 2.1 with PKCE and Dynamic Client Registration (RFC 7591) — the access token Claude receives is a normal Buddhi AI agent key under the hood, so it's subject to the exact same role-based scoping, citations, and audit logging as every other agent surface (see Authentication).

Revoking access

Disconnect the connector from Claude.ai's Settings, or revoke the underlying agent key from Gateway & Agents in the Buddhi AI dashboard — either one immediately and irreversibly stops the connection.

Example prompts

Once connected, try:

  • "What do we know about our rate-limiting policy for the public API?"
  • "Remember that our Q3 infrastructure budget is capped at $18k/month."
  • "What's our current MRR, and pull up any recent engineering decisions related to it."

The third exercises two tools in one turn (get_fact for the metric, recall for the related decisions) — a good way to confirm multi-tool sequencing works end to end.