Install mumo in Grok Bot.

Let your Bot run the install

Paste this into your Grok Bot:

Go to https://mumo.chat/install/grok-bot.md and follow the
instructions to connect mumo.

It drives the whole setup and hands you the Google sign-in at the one point it needs you.

Connect your Bot to mumo

Your Bot has its own computer and can mint your key directly so it never hits this chat.

Ask your Bot to open mumo.chat on its machine. You sign in with Google in its browser, it creates a key at /settings/api-keys, and it wires the MCP server.

That's a mumo sign-in, not a connector — a Gmail, Drive, or Calendar connector is a different thing and won't authenticate you here.

Prefer to handle the key yourself? Sign in to mumo via Google auth, create a key that starts with mmo_live_, then continue to step 02.

Add the mumo MCP server

Tell your Bot to add a remote MCP server (or paste the prompt below). It only needs the endpoint and a Bearer token:

Add an MCP server named mumo.
URL: https://mumo.chat/api/mcp
Transport: HTTP (Streamable HTTP)
Authorization: Bearer mmo_live_…

If you minted the key yourself, put it in the protected field your Bot offers — not in the chat transcript.

Once connected, the server sends short instructions for the deliberation loop (create_deliberation → wait_for_round → append_round). Smoke-test with list_models; a frontier-model list means you're live.

Remember, a key on an always-on agent is a live credential. It spends from your mumo balance and keeps working until you revoke it at /settings/api-keys. If teammates can message the Bot, treat that key as shared spend — give it its own key so revoking it costs you nothing else.

Run your first deliberation

Name mumo explicitly the first time so your Bot reaches for the panel instead of answering on its own:

Ask mumo to compare Postgres and MongoDB for our event store.
Include the constraints we discussed and tell me where the models disagree.

A panel takes roughly 15–120 seconds — the Bot fires create_deliberation, then polls with wait_for_round. A long wait is normal, not a hang.

You get back each model's full response — Claude, GPT, Kimi, DeepSeek, whichever panel was convened — plus a claim map of where they actually agree and diverge. From there your Bot can forward specific quotes into another round with append_round — challenge a claim, dig into a thread, or push back on the point everyone glossed over.