Connect Gemini CLI to Your Blog With MCP

Key takeaways

• Gemini CLI reads MCP servers from ~/.gemini/settings.json (user) or .gemini/settings.json (project).
• For a remote server use url for SSE or httpUrl for streamable HTTP, never both.
• Roami's Gemini endpoint is streamable HTTP (https://mcp.roami.xyz/mcp), so it uses httpUrl. Claude and Cursor use the SSE endpoint instead.
• Run /mcp inside Gemini CLI to see connection state and tools. Agents can draft; only you publish.

The short version of this setup lives on the Roami for Gemini page; this guide goes deeper on config and troubleshooting.

Where do I configure it?

Per the Gemini CLI MCP documentation (geminicli.com), servers go under the mcpServers object in settings.json. The user-level file is ~/.gemini/settings.json; the project-level file is .gemini/settings.json.

Which key do I use: url or httpUrl?

Use one, not both. Roami documents two endpoints on its developers page: https://mcp.roami.xyz/sse (SSE, for Claude and Cursor) and https://mcp.roami.xyz/mcp (streamable HTTP, for Gemini). So for Gemini:

{
  "mcpServers": {
    "roami": {
      "httpUrl": "https://mcp.roami.xyz/mcp"
    }
  }
}

If you prefer the command line, the Gemini docs show an equivalent form with --transport http:

gemini mcp add --transport http roami https://mcp.roami.xyz/mcp

The docs also list an optional timeout and headers; you don't need headers for the sign-in flow.

Check on your machine: if the connection shows DISCONNECTED, confirm you used httpUrl with the /mcp path, not url with /sse.

How does sign-in work?

For remote servers you can usually leave OAuth settings out. According to the Gemini docs, when a server returns 401 the CLI discovers the OAuth endpoints, may register a client dynamically, opens your browser for sign-in, and stores the tokens. If you need to re-authenticate, run /mcp auth to list servers that need it or /mcp auth roami for this one.

One catch: OAuth needs a local browser and a localhost callback. The docs say it will not work in headless environments, remote SSH sessions without X11 forwarding, or containers without browser support. Do the first sign-in on a machine with a browser. If you can't, Roami's developers page describes a self-serve MCP key (Developers → Create MCP key) sent as an x-api-key header; keep it out of any file you commit.

Step by step

How do I check the connection?

• Inside Gemini CLI, run /mcp. It lists each server with a state (CONNECTED, CONNECTING or DISCONNECTED), its configuration summary and its tools.
• From the shell, gemini mcp list shows connection errors.

Then try: "List my Roami projects." If you see projects, you are connected.

Prompts to try

Everything it writes arrives as a draft. Roami returns 403 Agents can draft and edit. Only you can publish. if any agent tries to publish, and Gemini is no exception. Gemini support was added on 13 September 2026. See Why AI Agents Should Not Publish.

Troubleshooting

Frequently asked questions

SSE or streamable HTTP? Use whichever the server documents for your client. For Gemini, Roami documents streamable HTTP.

User or project settings? User for a personal blog, project if a repo owns the blog.

Is a key required? No. OAuth sign-in works. Keys exist for developers and headless setups.

Try it

Start with Roami, add the entry above, and run /mcp. Then see Connect Cursor and Connect Claude.

Sources

• Gemini CLI: MCP servers, checked 18 Oct 2026.
• Roami developers page, checked 18 Oct 2026.