Quorum MCP
An MCP server and terminal where several models debate a question: Claude, GPT, Gemini, Grok and Ollama use one of seven methods
- 129 ★ on the repository
- Last commit
Extended access
Why this level
- The question and attached files go to several outside providers
- Agent CLI participants read the project and use the web and spend subscription quota
- API keys sit in the `~/.quorum/.env` file
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/quorumDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The question and attached files go to several outside providers
- Agent CLI participants read the project and use the web and spend subscription quota
- API keys sit in the `~/.quorum/.env` file
This is third-party code. Review the repository files before installing.
What it does
Quorum puts one question to several models and runs a structured discussion among them, after which one participant writes the final synthesis. There is a terminal version installed with `pip install quorum-cli`, and a plugin for Claude Code and Codex with the `quorum-mcp-server` MCP server. Participants can be the `claude`, `codex`, `agy` and `grok` agent CLIs, which run on your subscription without an API key, read the project read-only and use the web. Other participants come from `~/.quorum/.env`: OpenAI, Anthropic, Google, xAI, OpenRouter, LM Studio, `llama-swap`, custom endpoints and Ollama. They see only the question and the files passed. There are seven methods: `standard`, `tradeoff` for choosing between options, `delphi` for estimates, `advocate` for stress-testing a plan, `oxford` for a for-and-against debate, `brainstorm` and `socratic`. The MCP tools are `quorum_list_models`, `quorum_start`, `quorum_wait` and `quorum_check`. Transcripts are saved to `~/.quorum/runs/`.
Who it is for. Developers and team leads who want to hear several models before an architecture decision and get a summary of the disagreements.
Good fit when
- A choice between two architectures should go through the `tradeoff` method before you decide
- A release plan has to be stress-tested: the `advocate` method makes part of the models hunt for weak spots
- A time estimate is worth collecting from several models with the `delphi` method over several rounds
- You need an idea pool for a task and the `brainstorm` mode with several participants fits
- Local Ollama models need to sit next to cloud ones in one discussion
Not a fit when
- You need a quick answer: one run takes minutes and spends subscription quota
- Code cannot be shown to outside services: the question and files go to every chosen provider
- The BSL 1.1 license does not suit your project: its terms differ from open licenses such as MIT
Example request
Run a Quorum with the `tradeoff` method between Claude and Codex: monolith or modular monolith for our billing, attach the files from `docs/billing/`MCP
- Transport
- stdio
- Authentication
- key is optional
Environment variables
OPENAI_API_KEYsecret- OpenAI API key for participants from this provider
ANTHROPIC_API_KEYsecret- Anthropic API key for participants from this provider
GOOGLE_API_KEYsecret- Google AI Studio key for Gemini models
XAI_API_KEYsecret- xAI key for Grok models
Limitations
Python 3.11 and Node.js 18 are required, and the plugin also needs `uv` because the server starts through `uvx --from quorum-cli quorum-mcp-server`. The server always reads `~/.quorum/.env` and never the working project's `.env`. By default there are at most four participants, a turn takes up to 10 minutes, a whole run up to 30 minutes, and only one run goes at a time per project. The license is BSL 1.1. The old `quorum_discuss` interface is deprecated. The agent should start Quorum only on an explicit request, otherwise it spends subscription quota.
How to disable. Remove the plugin through `/plugin` in Claude Code or the Codex plugin manager. A server added without the plugin is removed with `claude mcp remove quorum`. The package is uninstalled with `pip uninstall quorum-cli`, while keys and transcripts stay in `~/.quorum/` until deleted by hand.
FAQ
How do I install Quorum in Claude Code?
Run `claude plugin marketplace add Detrol/quorum-cli`, then `claude plugin install quorum@quorum`. Without the plugin, install `pip install quorum-cli` and register the server with `claude mcp add quorum --scope user -- quorum-mcp-server`.
Do participants need API keys?
The `claude`, `codex`, `agy` and `grok` agent CLIs need no keys because they run on your subscription. Keys for OpenAI, Anthropic, Google and xAI models go into `~/.quorum/.env`, and Ollama is discovered automatically.
What does a participant look like in a request?
A participant is written as `provider:model@effort`, for example `claude:opus@high`, `openai:gpt-5.5@medium` or `ollama:qwen3:8b`. `quorum_list_models` returns the available options.
What about a long run?
Use `quorum_wait`: it polls the run in slices and is safe under a client's tool call timeout. A whole run is capped at 30 minutes.
Which discussion methods are there?
Seven: `standard`, `tradeoff`, `delphi`, `advocate`, `oxford`, `brainstorm` and `socratic`. The method is chosen to fit the question or named directly in the request.
Discussion
No questions or reviews yet.
Related
MCP serversbrainstorm-mcp runs debates among GPT, Gemini, DeepSeek, Claude and Ollama models in your editor and returns a three-point verdict
Plugin and MCP server that asks GPT, Gemini, Grok and local models to review a plan or hunt for bugs before code ships.
An MCP server that lets Claude Code, Codex and other agents call Gemini, GPT, Grok and local models for review, debugging and consensus
A local web service and MCP server that gathers several models to answer, peer review each other or debate as advisor personas.
A Claude Code plugin that asks several AI models the same question and shows their answers side by side with a synthesis of agreement and divergence
An MCP server that sends one prompt to several OpenAI, Anthropic, Gemini, Groq, DeepSeek and Ollama models at once and can draw a decision from them
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Quorum (this card) | no | key is optional | 129 | |
| LLM Council Skill | no | 501 | a model council skill without its own MCP server, while this is a server with seven methods and agent participants | |
| claude-council | no | 936 | a council plugin for Claude Code, while here agent CLIs, API models and local ones all take part |
README in short
The README presents Quorum as a war room in the terminal where models debate using formal methods. It goes on with the pip install, the plugin for Claude Code and Codex with its MCP tools and defaults, a source install for Linux, macOS and Windows, key setup in `.env` and discussion settings.