MCP Connector

IntelliRepo ships a remote Model Context Protocol (MCP) server, so you can use your knowledge base directly from AI assistants — Claude, ChatGPT, Claude Desktop, Cursor, Codex, Cline, and anything else that speaks MCP. Once connected, the assistant can search your collections and answer questions grounded in your documents, with citations.

There's nothing to copy or paste in the usual flow — you authorize the connection in your browser, the same way you'd connect any OAuth app.

Server URL: https://intellirepo.ai/api/v1/mcp

What the connector can do

Once connected, the AI assistant gets three tools:

You don't have to name the tools. Ask the assistant a question like "what's our PTO policy?" and it will call ask_collection on its own. You can also nudge it: "look it up in my IntelliRepo docs."

Connect from Claude (web)

1. In Claude, go to Settings → Connectors → Add custom connector.
2. Paste the server URL: https://intellirepo.ai/api/v1/mcp
3. Claude opens an IntelliRepo authorization page. If you're not signed in to IntelliRepo, sign in first.
4. On the consent screen, choose:
   • Workspace — the connection is tied to one workspace. To connect another, add the connector again.
   • Collections — all collections in that workspace, or a specific subset.
5. Click Approve. Claude stores the connection and the tools become available in your conversations.

Connect from ChatGPT

ChatGPT supports MCP servers through its custom connectors. In Settings → Connectors → Add, paste https://intellirepo.ai/api/v1/mcp. You'll be sent to the same browser authorization page above; on approval, ChatGPT registers the connection and the IntelliRepo tools become available in your conversations.

Connect from Claude Desktop / Cursor / Codex / Cline

These apps support the same browser-based flow — look for Settings → MCP (or Connectors), add a custom server, and paste https://intellirepo.ai/api/v1/mcp. You'll be sent to the same authorization page; no config file editing needed.

Older clients (config-file fallback)

Some older MCP clients can't do the OAuth flow and need a config entry. If yours is one of them, add something like this to its MCP config (exact key names vary by client):

{
  "mcpServers": {
    "intellirepo": {
      "url": "https://intellirepo.ai/api/v1/mcp"
    }
  }
}

If your client still can't authenticate, it predates remote-OAuth MCP support — upgrade it, or use the API + an API key instead.

Managing connections

Go to Settings → Integrations → AI Connectors in IntelliRepo to see every connected client, which workspace and collections each can access, and when it was last used. Click Revoke to cut off a connection immediately — its access token stops working on the next request, and the AI client will prompt you to reconnect if you want it back.

Usage & quotas

Each MCP tool call counts as one query against your plan, just like a chat message. AI agents typically make 3–10 tool calls per question (list collections → search → ask → maybe search again to verify a citation), so a connected agent can move through your monthly query allowance faster than the in-app chat. Your MCP usage shows up as its own line in Settings → Usage.

On trial accounts there's a hard cap of 200 MCP tool calls per day (resets at midnight UTC) as a guardrail; paid plans don't have that cap. There's also a per-connection burst limit (~100 calls per hour) — a runaway agent loop gets a 429 rather than draining your quota in seconds.

Security notes

• The connection authorizes the user who approved it. Anyone using that AI client account can query through the connection, attributed to that user — the same as an API key. Revoke from Settings if that changes.
• Tokens are scoped to one workspace (and optionally a subset of collections) — a connection can never reach data in another workspace or outside its collection allowlist.
• We use OAuth 2.1 with PKCE; the authorization code never lives in the consent page's JavaScript.
• Deleting a workspace or revoking a connection invalidates its tokens immediately.

Troubleshooting