CLAUDE COUNCIL MCP
Claude Council MCP plugs Council AI into every Claude surface — Desktop, Code, and Claude.ai via Connectors — and exposes a council of Claude Opus 4.7, GPT-5.5, Gemini 3.1 Pro, Grok 4.3, DeepSeek V4 and 22 more models as native MCP tools. One JSON config or one `claude mcp add` command, then Claude can call `council_query` and `council_query_with_rag` for cross-lab verification grounded in your own PDFs.
Claude Council MCP is Council AI's Model Context Protocol server purpose-built for the Claude ecosystem. Install it once in Claude Desktop (paste a JSON snippet into claude_desktop_config.json), Claude Code (run `claude mcp add --transport http council-ai https://mcp.council-ai.app/mcp`), or Claude.ai Connectors — and Claude can immediately call a council of 27+ frontier models from every lab: Claude Opus 4.7, Sonnet 4.6, GPT-5.5, GPT-5.4, o3, Gemini 3.1 Pro, Gemini 3.1 Flash-Lite, Grok 4.3, DeepSeek V4 Pro, Qwen3.6-Max, Mistral Large 3, Kimi K2.6 and more. The `council_query` tool fans your prompt across the council in parallel and returns a moderator-synthesized answer with a consensus score; `council_query_with_rag` does the same but first grounds every model in retrieved chunks from your personal Council library — the only Claude MCP that bundles personal RAG. Endpoint: https://mcp.council-ai.app/mcp. Auth: Bearer token from Settings → MCP. Tier: Ultra ($199.99/mo). Free to start trying the council in the web app — every premium model is open during the launch promo, no card required.
Claude Opus 4.7 is the strongest single model for nuance, writing, and long-context reasoning in May 2026. It's also confident in places where it shouldn't be — like every frontier model, it has training-data blind spots correlated across attempts. Asking the same question twice doesn't help; asking a different lab does.
A Claude council pairs Claude with models from OpenAI, Google, xAI, DeepSeek, Alibaba, Mistral, and Moonshot. When all of them agree, you have cross-lab consensus. When they diverge, that's the signal — the question is harder than it looked, or one model is wrong.
This is the same logic as a second opinion in medicine, peer review in science, or a code review across multiple senior engineers.
27+ frontier models total. The lineup refreshes as new releases ship.
Council AI Ultra ($199.99/mo) includes a hosted MCP server at https://mcp.council-ai.app/mcp. Add a small JSON snippet to your claude_desktop_config.json and the council appears as native tools — council_query, council_query_with_rag, library_search, get_models, get_usage.
In Claude Code, one CLI command:
claude mcp add --transport http council-ai https://mcp.council-ai.app/mcp \
--header "Authorization: Bearer csa_YOUR_TOKEN_HERE"Generate your token in Settings → MCP tokens. Full setup guide at /mcp.
Claude Council MCP is Council AI's hosted Model Context Protocol server purpose-built for Claude Desktop, Claude Code, and Claude.ai Connectors. The endpoint is https://mcp.council-ai.app/mcp. Once installed, Claude can call six native tools — council_query, council_query_with_rag, library_search, library_list, get_models, get_usage — to invoke a council of 27+ frontier AI models (Claude Opus 4.7, GPT-5.5, Gemini 3.1 Pro, Grok 4.3, DeepSeek V4, Qwen3.6-Max, Mistral Large 3, Kimi K2.6 and others) directly from inside Claude. Available on the Ultra tier ($199.99/mo).
Generate a Personal Access Token in Council Settings → MCP, then open claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/claude_desktop_config.json, Windows: %APPDATA%\Claude\claude_desktop_config.json) and add a mcpServers entry with url "https://mcp.council-ai.app/mcp" and an Authorization: Bearer csa_YOUR_TOKEN header. Restart Claude Desktop and the council appears as native tools.
From your project root, run: `claude mcp add --transport http council-ai https://mcp.council-ai.app/mcp --header "Authorization: Bearer csa_YOUR_TOKEN_HERE"`. The server registers at user or project scope; Claude Code can then call council_query and council_query_with_rag inline during coding sessions for cross-lab review or RAG-grounded design feedback.
No — Claude is one of 30+ models in the council, alongside GPT-5.5, Gemini 3 Pro, Grok 4.3, DeepSeek V4, and others. The point is to get cross-lab verification, not just more Claude. Council AI's moderator quantifies cross-model agreement so you know when the labs converged on the same answer.
No. Council AI is a separate product that includes Claude Opus 4.7, Sonnet 4.6, and Haiku 4.5 in its own subscription. If you do have Claude Pro or Claude Code, the Ultra-tier MCP server lets you invoke the entire Council AI lineup from inside those tools too.
You skip the copy/paste, get parallel streaming, and an AI moderator that scores agreement and synthesizes a final answer naming which models said what. On Ultra, you also get personal RAG — every model in the fan-out first retrieves from your own PDFs and Word docs.
Yes. Once you've added the MCP server with `claude mcp add`, Claude Code can call `council_query` for cross-lab review whenever it wants a second opinion, or `council_query_with_rag` to ground that review in your codebase notes and architecture docs.
Free to start — every premium model open during the launch promo. Starter $4.99/mo for persistent memory and higher limits. Pro $59.99/mo for daily heavy use. Ultra $199.99/mo unlocks the MCP server (Claude Desktop / Claude Code integration) and a personal RAG library across all models.