MCP quickstart
Enable the MCP server and connect your AI client over OAuth.
Last verified: 31 July 2026
The MCP server requires the Full tier.
Enable the server
In the web app, go to Settings → Developers → MCP and enable the server.
Copy your server URL
The settings page shows your MCP endpoint – the /mcp path on your Remly application host. Copy it from there.
There is no token to copy. Authentication happens through OAuth in the next step.
Add the server to your client
The settings page generates the exact configuration for each supported client. The shapes below are what it produces; use the page as the authority if they differ.
Claude.ai and Claude Desktop
Settings → Connectors → Add custom connector, and paste your Remly MCP URL.
Claude Code
claude mcp add --transport http remly <your-mcp-url>Then run /mcp inside Claude Code and pick Remly to start the OAuth flow.
Cursor
Settings → MCP → Add new MCP server, and paste the JSON the settings page gives you. It points at your MCP URL over HTTP.
ChatGPT
Settings → Connectors → add a connector pointing at your MCP URL.
Any other MCP client
Point it at your MCP URL using Streamable HTTP transport with OAuth 2.1. SSE is not supported and will fail.
Approve the client
The first time the client connects, it opens a browser window where you sign in to Remly and approve access. The client then stores its own OAuth credentials.
You will not paste a bearer token anywhere in this flow. If your client is asking for one, it is either configured for the wrong transport or following outdated instructions.
Try it
Ask your client to search your Remly content. It should call search, get back results, and be able to call get_item on any result to read it in full.
Revoking a client
Disconnect the client from Settings → Developers → MCP, or disable the server entirely to cut off every client at once. Revoking invalidates that client's OAuth credentials – it will need to go through browser approval again to reconnect.
Because a connected client can search and open any document in your own indexed knowledge base, revoking is the right move any time you stop trusting a client or a machine it runs on.
Usage logs
The same settings page shows recent calls: time, client, tool, and status. Query text and document contents are not logged.
Limits and errors
get_item is capped at 3,000 reads per day, and get_context is only exposed when the Context Layer is enabled for your account. See usage limits.
| Error | Cause |
|---|---|
| Auth failures on connect | Wrong transport (SSE instead of Streamable HTTP), or an expired grant. Reconnect and approve again. |
| Tools missing from the client | get_context only appears when enabled and entitled. search and get_item should always be present. |
| Tool not found | The client is configured for the removed remly.ask or remly.list_recent tools. |
| Cap reached | Daily allowance exhausted. It resets in the next window. |
| Server unavailable | The MCP server is disabled in settings, or your subscription is not on the Full tier. |