mumla

Connect your AI agent to Mumla — the governed egress layer.

Your agents can think. Now they can talk — governed. Every message an agent sends through Mumla is held for a human's call when it matters, delivered on the channels you allow, and recorded as a signed receipt. Your agent holds a permission slip; Mumla holds the keys.

Use it

npx @mumla/cli

That's the whole thing. It opens app.mumla.ai so you can sign in and create an agent, then wires that agent into Claude Code so it can send through your Mumla endpoint. Then ask your agent to send an email — and watch Mumla hold it for your approval before anything leaves.

npx @mumla/cli connect is the same command, spelled out.

What it actually does

• Opens the browser to app.mumla.ai (sign in → create an agent → copy its token).
• Wires Claude Code: claude mcp add mumla https://api.mumla.ai/mcp --header "Authorization: Bearer <token>".
• Any MCP client that accepts a bearer header works — the CLI prints the exact command so you can paste it anywhere.

The token you paste is handed straight to claude mcp add. This CLI never writes it to disk.

Requirements

• Node 18+
• Claude Code for automatic wiring (optional — the CLI prints the manual command for any other MCP client).

License

MIT