A local MCP server brings an AI council into coding agents
llm-council-mcp-server is a local tool for getting more than one AI-style view on a hard question before turning the result into one answer. It works as an MCP server inside such as Claude Code, Codex CLI, and . The free internal council mode does not need an OpenRouter key; the current agent imitates five viewpoints: a pragmatist, an architect, a skeptic, a clean-code reviewer, and a product or UX thinker.
The full multi-model mode uses OpenRouter and follows three steps: separate answers from different models, anonymous review of those answers, and a final synthesis by a . It can run from PyPI through uvx, so there is no need to clone the repository. A browser setup screen lets users change the API key, models, , and .
The docs and setup screen are available in English and German, and the tool is still early and mainly focused on Mac for now.
Key points
- The tool is a local MCP server for Claude Code, Codex CLI, and .
- The free internal council mode works without an OpenRouter key.
- The full multi-model mode uses separate answers, blind review, and final synthesis.
- It can run from PyPI with uvx, without cloning the repository.
- A browser setup screen controls API keys, models, , and temperature.