MCP server
Connect Remly to Claude, Cursor, ChatGPT, and any other MCP client over OAuth.
Status: Available (Full tier) · Last verified: 31 July 2026
Remly exposes your indexed content to any Model Context Protocol client, so an AI client you already use can search and read your content with your permissions.
The contract
| Endpoint | The /mcp path on your Remly application host. The exact URL is shown in Settings → Developers → MCP – copy it from there rather than assembling it yourself. |
| Transport | Streamable HTTP. SSE is not supported. |
| Authentication | OAuth 2.1 through Clerk. You approve the client in a browser and it receives its own credentials. |
| Tier | Full tier only. |
| Tools | search, get_item, and – when enabled – get_context. See the tool reference. |
There is no token to copy and no token in the URL. If a setup guide anywhere tells you to paste a static bearer token from your Remly settings, it is out of date. Authentication happens through a browser approval step, and the client stores its own credentials afterwards.
What a connected client can reach
A connected client can search and open any document in your own indexed knowledge base. There is no per-source or read-only limit yet, so only connect clients you trust.
Access is scoped to your organisation and to your own document permissions. A client sees exactly what you can already see in the web app, and nothing beyond it – not other people's documents, and not other organisations' content.
What the missing per-source limit means in practice: you cannot authorise a client for one connector and exclude another. If you have connected email, meeting transcripts, or calendars, an authorised client can search across all of them.
Usage logging
Settings → Developers → MCP shows recent calls with metadata: time, client, tool called, and status. Query text and document bodies are not stored in those logs.
Where to start
- Quickstart – enable the server and connect your client
- Tool reference – parameters, defaults, and what comes back
- Examples – workflows worth stealing
Removed tools
Earlier documentation described remly.ask and remly.list_recent. Neither exists. The current server exposes search, get_item, and conditionally get_context. If a client is configured against the old names, it will fail.