Guide · Clients

Wire Gemini to Munin.

The Gemini CLI reads remote MCP servers from settings.json. Add Munin and Gemini gains every tool the calling key has scope for — KB hybrid search, CRM lookups, conversation drafts, CMS writes.

1 · Mint an API key

From the dashboard, go to Settings → API keys. Pick scopes that match what Gemini should be allowed to do. The token starts with mn_admin_ and is shown once.

2 · Add the server (CLI)

The fastest way is the one-shot CLI command — it writes the entry into the project (or user-level) settings for you:

shellwrites ~/.gemini/settings.json

gemini mcp add --transport http munin http://localhost:3001/mcp \
  -H "Authorization: Bearer mn_admin_…"

Or edit settings.json directly

Gemini CLI reads ~/.gemini/settings.json (user scope) or .gemini/settings.json in the project root. Merge Munin into mcpServers. Use httpUrl for the streamable HTTP transport (url is reserved for SSE).

settings.jsonmerge with existing servers

{
  "mcpServers": {
    "munin": {
      "httpUrl": "http://localhost:3001/mcp",
      "headers": {
        "Authorization": "Bearer mn_admin_…"
      }
    }
  }
}

3 · Verify

Run gemini mcp list— it prints every connected server and the tools it discovered. If Munin doesn’t show, re-run with --verbose to see the handshake error. Then prompt Gemini in a session with “List the Munin tools you can call” as a smoke test.

If the list is empty, the most common causes are a stale config (restart the gemini shell), a wrong path (it’s /mcp, not /v1/mcp), or a revoked token.

4 · Tighten scope

Mint a separate key per workstation, with the smallest scope that lets the model do its job — read-only KB is plenty for a Q&A assistant, full conv:write is needed only if Gemini should actually send replies.