Deliberation plugin
Plugin and MCP server that asks GPT, Gemini, Grok and local models to review a plan or hunt for bugs before code ships.
- 183 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Code snippets and plans are sent to third-party model providers such as OpenAI, Google, xAI and OpenRouter.
- The Gemini expert can modify project files in write mode.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/deliberationDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Code snippets and plans are sent to third-party model providers such as OpenAI, Google, xAI and OpenRouter.
- The Gemini expert can modify project files in write mode.
This is third-party code. Review the repository files before installing.
What it does
Deliberation gives a coding agent second opinions from other models: GPT, Gemini, Grok, local ones through Ollama and LM Studio, and more than 400 models via OpenRouter. Each expert has a specialty: architect, code reviewer, security analyst, debugger, writing editor and others, eight in total. An expert analyzes read-only or edits in write mode, and only Gemini can write. In Claude Code the plugin adds `/consensus` and `/ask-all`: the first runs reviewers in rounds until they agree, the second collects independent verdicts and a disagreement matrix. Auto-routing detects when a task is worth handing to an expert. The agent paraphrases the answer and never returns it raw. The same MCP server runs in Codex, Cursor, Kiro, Antigravity and OpenCode, and connects separately in Claude Desktop, Windsurf and VS Code. An opt-in local dashboard on 127.0.0.1 draws each run as a timeline.
Who it is for. Developers who do not trust a single model for review and design and want decisions cross-checked with models from other vendors.
Good fit when
- An architecture plan is ready and needs independent reviewers before implementation.
- Tests fail for no clear reason and you want a debugger running on another model.
- Two solution variants are contested and the team wants several verdicts with a disagreement matrix.
- A cheap local Ollama model should join as an adviser beside the main one.
Not a fit when
- Code must not be sent to third-party model providers.
- One review from the main model is enough and extra calls are unwanted.
- You have no subscription or key for any of the providers.
Example request
Run /consensus on the database migration plan and show where the reviewers still disagree after round five.Limitations
Each expert goes through its own provider: GPT and Gemini call the `codex` and `agy` command-line clients, which you log into separately, Grok uses `XAI_API_KEY`, and OpenRouter uses `OPENROUTER_API_KEY` plus a model list in `~/.config/deliberation/config.json`. Code snippets and your plan are sent to those providers. Write mode exists only for Gemini. One-click install buttons cannot carry keys, so the `env` block is added by hand. The local dashboard is off by default and stores only metadata.
How to disable. In Claude Code run `/plugin uninstall deliberation@antonbabenko`. For the standalone MCP server, remove the `deliberation` entry from the client config and clear any keys in `env` by hand.
FAQ
Which models can I connect?
GPT, Gemini and Grok, local models through Ollama and LM Studio, and any OpenRouter model including Qwen, Kimi and DeepSeek. Any subset is enough.
How does `/consensus` differ from `/ask-all`?
`/consensus` repeats review rounds until the experts agree. `/ask-all` collects independent verdicts and asks the models to critique each other.
Does it work only in Claude Code?
No. Native plugins exist for Codex, Cursor, Kiro, Antigravity and OpenCode, and the standalone npm MCP server fits any other client.
Who can change code?
All experts advise, but only Gemini can edit files in write mode. The others stay read-only.
How do I enable the local dashboard?
Add `"dashboard": { "enabled": true }` to the config and open `/deliberation:dashboard`. The page is read-only and listens on 127.0.0.1.
Discussion
No questions or reviews yet.
Related
PluginsA 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 lets Claude Code, Codex and other agents call Gemini, GPT, Grok and local models for review, debugging and consensus
A Claude Code plugin that runs one task across several external models and surfaces disagreements before you ship
brainstorm-mcp runs debates among GPT, Gemini, DeepSeek, Claude and Ollama models in your editor and returns a three-point verdict
A plugin that syncs Claude Code, Codex CLI and Antigravity on one project and lets them hand work to each other.
An MCP server and terminal where several models debate a question: Claude, GPT, Gemini, Grok and Ollama use one of seven methods
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Deliberation (this card) | no | 183 | ||
| PAL MCP | no | key is optional | 11,839 | also links an agent to several models; here eight specialist experts, round-based consensus and a local run dashboard |
| Consensus MCP Server | yes | OAuth | 31 | focuses on consensus; here consensus is joined by expert roles, a write mode and native plugins for six clients |
README in short
The project wires several providers into an MCP host so the user only writes the prompt. A set of eight experts, consensus commands and auto-routing ship in a single plugin. The author also publishes the standalone `@antonbabenko/deliberation-mcp` package and an entry in the official MCP registry.