Install mumo in OpenClaw.

Generate your API key

Sign in, create a mumo API key, and copy the value that starts with mmo_live_. You'll put it in OpenClaw's env file in the next step — the MCP config only references it.

Register the mumo MCP server

The key goes in ~/.openclaw/.env. Create the file with owner-only permissions before the key goes in, then open it in an editor and add a line MUMO_API_KEY=mmo_live_… with your key. Use an editor rather than an echo >> one-liner — a shell command carrying the key lands in your shell history:

touch ~/.openclaw/.env && chmod 600 ~/.openclaw/.env

OpenClaw stores outbound MCP servers in ~/.openclaw/openclaw.json under mcp.servers.<name>. Register mumo with the CLI exactly as written — the header references ${MUMO_API_KEY}, which OpenClaw resolves from the environment or ~/.openclaw/.env when it connects, so the registered block never holds a live credential:

openclaw mcp set mumo '{
  "url": "https://mumo.chat/api/mcp",
  "transport": "streamable-http",
  "headers": {
    "Authorization": "Bearer ${MUMO_API_KEY}"
  }
}'

openclaw mcp show mumo prints what was saved. That reads config only — it does not open a connection, so the real check is the free list_models call in step 05. If the variable is missing, OpenClaw warns at startup naming mcp.servers.mumo.headers.Authorization. To rotate your key, update .envand restart OpenClaw; the registered block doesn't change.

Install the mumo skill

Registering the MCP server gives the agent the tools, but not the discipline to use them well. The skill teaches OpenClaw the deliberation loop, claim-map reading, snippet doctrine, and how to verify a session was actually created. Without it, the agent has tools without a manual.

The fastest path is via ClawHub, OpenClaw's skill registry:

openclaw skills install mumo

That pulls mumo from ClawHub into your active agent workspace's skills directory (--agent <id> targets a different workspace).

If you'd rather pin the skill at a fixed path — or track the source repo directly to pull main ahead of registry releases — clone instead:

git clone https://github.com/mumo-chat/mumo-openclaw \
  ~/.openclaw/skills/mumo

Either way, the skill ships the canonical SKILL.md, four cognitive-shape playbooks (contested decision, design review, uncertainty expansion, red team), and reference docs for claim-map reading, snippet doctrine, model selection, and synthesis.

Restart OpenClaw

Fully exit and restart OpenClaw so it picks up both the new MCP server registration and the new skill. The coding and messagingtool profiles expose configured MCP servers by default; if you're on the minimal profile, MCP tools stay hidden.

After restart, the mumo tools surface as mumo__create_deliberation, mumo__wait_for_round, mumo__append_round, mumo__get_session, mumo__share_session, mumo__list_sessions, mumo__list_models, mumo__get_credit (OpenClaw uses the <server>__<tool> double-underscore convention).

Verify the skill loaded and the key resolves

In your OpenClaw session, run:

/skill mumo

You should see the mumo skill listed as ready with the description starting "Multi-model deliberation via mumo's MCP server…". Then smoke-test auth with a free call:

Ask mumo to list available models.

That runs mumo__list_models, which authenticates against the server and costs nothing — no deliberation is started. A model catalog back means the key resolved and the server accepted it. An auth error means MUMO_API_KEY is unset or wrong in ~/.openclaw/.env.

Run your first deliberation

Name mumo explicitly the first time so OpenClaw routes through the panel. The skill will guide the agent through the create→wait→read→snippet loop and teach it to verify the session actually fired.

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

If the agent claims a deliberation succeeded but doesn't produce a UUID-format session ID, the call didn't fire — confirm with mumo__list_sessions or check mumo.chat. The skill teaches this discipline; this is the user-facing reminder.