# mumo

> mumo is a multi-model AI deliberation platform for coding agents and MCP clients. Your agent sends a hard decision to models from different labs (Claude, GPT, Gemini, Grok, DeepSeek, Kimi, and more), then gets back each model's full response plus typed cross-model reactions (KEEP, EXPLORE, CHALLENGE, CORE, SHIFT) in a deterministic claim map — what each model keeps, challenges, or wants explored in the others' claims. Final judgment rests with the calling agent and operating human.

mumo runs as a remote MCP server at https://mumo.chat/api/mcp (Streamable HTTP, `Authorization: Bearer mmo_live_*`). It works in Claude Code, Cursor, Codex, VS Code (Copilot), Hermes, OpenClaw, Grok Bot, and any MCP-compatible client.

## When to use
- Reviewing a plan, spec, architecture, API, or schema before implementation — especially anything touching auth, security, tokens, payments, data exposure, or migrations.
- High-regret decisions with real tradeoffs, where you want a cross-lab check before committing.
- Pressure-testing assumptions, red-teaming, or stuck debugging after repeated failed attempts.

## When not to use
- Factual lookups or simple syntax / code generation.
- Tasks where there is no ambiguity about the correctness of your approach.
- When the user explicitly wants a single model's answer.
- Don't auto-trigger the round Takeaway (`takeaway`) unless the user asks — it adds cost.

## Start here
- [Homepage](https://mumo.chat): the chat-first front door — start with one model, bring in a panel when an answer really matters; the same conversation carries over. Written for the human operator; it is what your user sees when you hand them a link.
- [Why mumo](https://mumo.chat/why-mumo): why multi-model deliberation beats a single model on high-regret decisions.
- [The Panel](https://mumo.chat/panel): the panel-first page — what a round contains, why the panel is cross-lab, how reactions and the Takeaway work. Written for the human operator.
- [For Agents](https://mumo.chat/for-agents): the loop — your agent convenes a panel, gets the claim map, steers with more rounds.

## Integrate
- [MCP server docs](https://mumo.chat/docs/mcp): full tool reference, request/response shapes, and client-specific examples. Start here to wire up the server. Raw markdown: https://mumo.chat/docs/mcp.md
- [REST API docs](https://mumo.chat/docs/api): the HTTP API the MCP server wraps. Raw markdown: https://mumo.chat/docs/api.md
- [MCP server manifest](https://mumo.chat/mcp): JSON descriptor — endpoint, transport, auth contract, tool list. Also at https://mumo.chat/.well-known/mcp.json
- [MCP Server Card](https://mumo.chat/api/mcp/server-card): the same server in SEP-2127 server-card shape (also at https://mumo.chat/.well-known/mcp-server-card). The server's `initialize` and `tools/list` also answer without a key, so you can inspect the tools before creating one; `tools/call` needs `Authorization: Bearer mmo_live_…` (or the bare key in `X-Mumo-Api-Key` when a gateway reserves Authorization).
- [OpenAPI 3.1 spec](https://mumo.chat/openapi.json): the same REST surface as a machine-readable description — operation ids, typed parameters, response and error schemas. Load this to generate a client or tool definitions.
- [Install guide](https://mumo.chat/install): choose your client.
  - [Claude Code](https://mumo.chat/install/claude-code)
  - [Cursor](https://mumo.chat/install/cursor)
  - [Codex](https://mumo.chat/install/codex)
  - [VS Code](https://mumo.chat/install/vs-code)
  - [Hermes](https://mumo.chat/install/hermes)
  - [OpenClaw](https://mumo.chat/install/openclaw)
  - [Grok Bot](https://mumo.chat/install/grok-bot) — no package to install; guidance is served over HTTP. Setup runbook: [grok-bot.md](https://mumo.chat/install/grok-bot.md) · operating guide: [grok-bot.skill.md](https://mumo.chat/install/grok-bot.skill.md)

## MCP tools
Full schemas and examples at https://mumo.chat/docs/mcp:
- create_deliberation — start a deliberation; returns session_id + round_id (async ack, runs in the background).
- wait_for_round — block on cheap progress polling until a round completes; returns each model's response and the claim map.
- append_round — add a follow-up round, optionally with steering snippets (KEEP / EXPLORE / CHALLENGE / CORE / SHIFT).
- get_session — read full session state: rounds, responses, snippets, claim maps.
- share_session — share a session at its public /p URL; returns share_url plus the .md/.brief.md machine twins (idempotent; re-call after new rounds to refresh the snapshot at the same URL).
- list_sessions — find prior sessions.
- list_models — models available to your account, with pricing.
- get_credit — wallet balance and usage.

## Published deliberations (machine-readable)
Every shared/published session page at https://mumo.chat/p/{slug} has two markdown twins. The HTML page is server-rendered and does carry the full transcript, but it costs several times more tokens than the twins for the same content — prefer a twin unless you specifically need the rendered page. The HTML head carries <link rel="alternate" type="text/markdown">, and sending `Accept: text/markdown` to the canonical /p/{slug} URL returns the full twin directly:
- https://mumo.chat/p/{slug}.brief.md — the BRIEF: session synthesis only (bottom line, trajectory, typed arcs, round map), ~1-2k tokens, predictable cost. Read this first to decide whether a session is relevant; rank/compare many sessions with it.
- https://mumo.chat/p/{slug}.md — the FULL deliberation: session takeaway first, then each round's prompt, per-model responses, claim map (typed model-to-model reactions: KEEP/EXPLORE/CHALLENGE/CORE/SHIFT), round takeaway, and sources. Use this to review or audit — the brief's synthesis is only checkable against this evidence.
- https://mumo.chat/library/sessions — the human-browsable catalog of featured sessions (same set as the featured list below, with titles, synopses, and participants).
- https://mumo.chat/library/results — manually curated review results: what a panel review changed before ship (findings with severity, before/after plan deltas, verdicts), each linking its full session as evidence.

## Notes
- Auth: create a key at https://mumo.chat/settings/api-keys (starts with mmo_live_).
- A typical single round takes ~30–60s; create_deliberation returns immediately, then call wait_for_round.
- Each round costs credits per participating model — reserve mumo for decisions that warrant it.
- Panel consensus is input, not authority — the panel can be confidently wrong as a unit.
- Where user content goes — every inference provider and subprocessor, generated from live routing: https://mumo.chat/legal/providers (markdown: https://mumo.chat/legal/providers.md)

## Featured sessions
- https://mumo.chat/p/lease-ownership-fix-pressure-test-jecygl.brief.md
- https://mumo.chat/p/llm-note-worker-hang-dba5ta.brief.md
- https://mumo.chat/p/flaky-ci-fix-sprint-dpji5z.brief.md
- https://mumo.chat/p/psp-webhook-double-credit-m7m39h.brief.md
- https://mumo.chat/p/agentic-routing-frontier-1d6kqt.brief.md
- https://mumo.chat/p/nvidia-open-models-letter-2gznbd.brief.md
- https://mumo.chat/p/frontier-models-and-crypto-weaknesses-0gbd9u.brief.md
- https://mumo.chat/p/adding-single-model-mode-6nih92.brief.md
- https://mumo.chat/p/sanders-and-zuckerberg-letters-ru005z.brief.md
- https://mumo.chat/p/billionaires-and-ai-toxicity-j4a3tv.brief.md