MCP Connect menu
Offer your hosted MCP server to readers from your docs. A Connect menu in the docs navbar adds it to Cursor, VS Code or Claude Code, so a reader goes from your docs to an agent that can use your API without copying anything by hand.
It's off by default. The hosted MCP server is live for every project either way, and advertising it is your call.
Turn it on
Open Integrations in the project sidebar, choose MCP server in your docs under Developer tools, and press Turn on. The menu appears on every docs page straight away, with no publish needed.
The same switch is in API Studio's Docs settings, under Browser agents, as Offer your MCP server in the docs.
What readers see
Connect opens a short menu:
| Item | What it does |
|---|---|
| Add to Cursor | Installs the server in Cursor in one click. Shown when sign-in is on |
| Add to VS Code | Installs the server in VS Code in one click. Shown when sign-in is on |
| Copy Cursor config | Copies the server's Cursor config. Shown in place of the one-click install when sign-in is off |
| Copy VS Code config | Copies the server's VS Code config. Shown in place of the one-click install when sign-in is off |
| Copy Claude Code command | Copies a claude mcp add --transport http ... line to paste in a terminal |
| Copy MCP server URL | Copies the address, for any other MCP client |
The address is your custom MCP domain while it's serving, and Octri's address for the server otherwise: the same one the dashboard hands out.
Credentials
What a config asks for depends on how the server signs people in:
- Sign-in on. The configs and the command carry no credential. Each reader signs in when their client first connects, which is why one-click installs are offered only then.
- Sign-in off. The configs and the command include the headers your API's authentication needs, with a placeholder for the reader's own key. Nobody's real key is ever in the menu.
See hosted MCP server for sign-in, and installing for what readers do with each config.
Related
Browser agents that read the docs page itself, rather than connecting a client, are covered in llms.txt and browser agents.