claude.ai is a hosted client: it can’t run a local process, so it can only reach a remote Model Context Protocol (MCP) server, and moodle mcp connect doesn’t write a config file for it — you add it through claude.ai’s own settings instead. Official references: Getting started with custom connectors using remote MCP and Third party connectors with remote MCP, both fetched 2026-09-21.

​
Prerequisites

  • A deployed Moodle Worker (moodle mcp deploy) or a deployed Ed Discussion Worker, or use Ed’s hosted endpoint.
  • For Moodle: nothing more, since claude.ai authenticates through OAuth pairing.
Nothing is approved on your Worker until you open a pairing window yourself:
This prints your connector URL and a one-time pairing code, valid for ten minutes and one approval.In claude.ai:
  1. Go to Customize → Connectors (Team/Enterprise owners: Organization settings → Connectors first, then members connect from Customize → Connectors).
  2. Click Add custom connector.
  3. Enter the connector URL moodle mcp pair printed.
  4. Leave authentication on its default OAuth detection — don’t switch to No sign-in or add a Request headers entry; the Worker only accepts OAuth from claude.ai.
  5. Click Add, then Connect. claude.ai opens the Worker’s approval page; sign in and enter the pairing code.
From then on, claude.ai stays signed in on its own, and you don’t need to repeat pairing unless you revoke it.

​
Verify the connection

Open the ”+” menu in a chat and check Connectors — the server should be listed and toggled on. Enable it for the conversation and ask a question that requires a tool call; Claude will show a permission prompt naming the tool the first time it’s used. Only click Always allow for a connector you trust.