Mumo MCP Server

mumo — Claude Code + Cowork plugin

Multi-model responses + cross-model reactions. Want more rounds? Context carries automatically. Stop when you have what you need.

Claude, GPT, Gemini, Grok, Qwen, GLM, Kimi in parallel. For contested decisions — architecture, plan review, pricing, strategy — where a single model might be confidently wrong.

Works in both Claude Code (developer audience) and Claude Cowork (knowledge-worker audience). Same plugin, same skill, different distribution marketplaces. For Cursor, see mumo-chat/mumo-cursor. For VS Code (GitHub Copilot), see mumo-chat/mumo-vscode.

When to use

Reach for mumo when:

• Architecture and design reviews — "Postgres or MongoDB for our event store?" "Is this migration safe under concurrent writes?"
• Plan and pre-launch pressure tests — "What would we regret 6 months in?" "Where's the biggest risk in this spec?"
• Contested product or pricing decisions — "Should this feature be paid or free?" "Which of these positionings actually converts?"
• Strategy with real cost — anywhere a confidently-wrong single model could push a team in the wrong direction for a week.

The bundled skill auto-triggers on this category and stays quiet on factual lookups, routine refactoring, and code generation from a clear spec.

What's in the box

• MCP server — https://mumo.chat/api/mcp, six tools: create_deliberation, append_round, get_session, list_sessions, list_models, get_credit
• Auto-triggering skill — skills/mumo/SKILL.md tells the agent when to reach for the panel so you don't have to

Install — Claude Code

Inside Claude Code:

/plugin marketplace add anthropics/claude-plugins-official
/plugin install mumo

Install — Claude Cowork

Open Claude Desktop → Cowork tab → Customize → Browse plugins → search for mumo → Install. Or browse the full catalog at claude.com/plugins.

Note: Claude Code and Cowork have separate plugin panels backed by different marketplaces. Installing in one surface does not auto-install in the other — install in each separately.

Install — VS Code (GitHub Copilot)

Install the mumo extension from the Visual Studio Marketplace. It registers the MCP server natively and stores your key in VS Code's SecretStorage — no MUMO_API_KEY env-var required.

For the full story and manual-install alternatives, see mumo-chat/mumo-vscode.

The auto-triggering skill from this repo is not bundled in the VS Code extension — Copilot's agent mode routes to the MCP server organically when you mention mumo or ask for a panel. For the explicit skill experience, use Claude Code or Cowork.

API key setup

Claude Code and Cowork both need a mumo API key:

1. Sign up at https://mumo.chat and create a platform key at Settings → API Keys (keys start with mmo_live_)
2. Export it in your shell:

   export MUMO_API_KEY=mmo_live_YOUR_KEY_HERE
   

3. Restart Claude Code or Claude Cowork

Using the panel

Invoke the skill explicitly if the agent doesn't reach for it on its own:

• "Run this by a mumo panel"
• "Get me a second opinion from mumo on…"
• "Ask mumo about…"

Try it first

Ask Claude Code or Cowork something the panel will want to engage with:

Postgres or MongoDB for our event store given 50k events/day, a Postgres-experienced team, and a 3-month runway? What would we regret 6 months in?

The first round returns each model's prose plus a cross-model claim map showing where the panel agrees and where it splits. You can stop there, or append_round with typed snippets (KEEP / EXPLORE / CHALLENGE / CORE / SHIFT) to steer further.

See mumo.chat/mcp for the tool reference, the deliberation loop, and prompt patterns. The canonical trigger language lives in skills/mumo/SKILL.md — that's what Claude Code loads.

Links

• Product — https://mumo.chat
• MCP reference — https://mumo.chat/docs/mcp/reference
• REST API — https://mumo.chat/docs/api
• Cursor plugin — https://github.com/mumo-chat/mumo-cursor
• VS Code extension — https://github.com/mumo-chat/mumo-vscode
• Issues — https://github.com/mumo-chat/mumo-mcp/issues

License

MIT