Bring the Council to Your AI

When a decision actually matters, one AI model isn't enough. Different models reach different conclusions, miss different things, and carry different blind spots — and on a hire, a pricing change, a contract term, or an architecture call, you want to know where they agree and where they don't before you commit.

That's what ModelCouncil does. It sends your question to Claude, GPT, Gemini, and Grok at the same time and synthesizes their answers into a single Decision Board: a clear recommendation, the key reasons behind it, the risks, where the models reached consensus, where they disagreed, and the rare insight only one of them surfaced. One question in, one structured read on the decision out.

Until now, getting that read meant coming to ModelCouncil — opening the app, setting up the question, re-supplying context you'd often already built somewhere else. Today that changes.

ModelCouncil now runs as an MCP server. You can convene the council from inside the AI tool you already work in — Claude Desktop, Claude.ai on the web, ChatGPT, Cursor, Claude Code — without switching tabs or repeating yourself. You ask; the council answers; you stay in your workflow.

How it works

Model Context Protocol is an open standard for connecting AI assistants to external tools. Connect ModelCouncil as one of those tools, and your assistant can hand a question to the full council and bring back a structured answer.

In practice, it looks like this. You're working through a problem in Claude. You hit a question that matters, and you say:

"Run this past the council and tell me where they disagree."

Claude calls ModelCouncil. Your question goes to all four models in parallel, gets synthesized into a Decision Board, and comes back into your conversation. The deliberation happens behind the scenes; the answer arrives where you already are.

The point isn't a second opinion from one more model. It's a structured read on where four frontier models agree and where they split — and the splits are usually where the real decision lives.

The five tools

Once connected, your AI assistant has five ModelCouncil tools available. You don't invoke these directly — you just talk to your assistant in plain language, and it picks the right one.

• list_projects — finds your existing ModelCouncil projects so a query lands against the right context.
• query_council — the standard query. Sends your question to the council and returns a full Decision Board in a single call.
• query_diamond_mode — the deep one. Runs Diamond Mode, where the models review and revise each other's answers across two rounds before final synthesis. It returns immediately and finishes in the background, so a long deliberation never blocks your conversation.
• get_query_status — checks on a Diamond query in flight, or pulls back a completed Decision Board by ID.
• create_project — spins up a new project when none of your existing ones fit the topic.

Every council query run this way still appears in your ModelCouncil dashboard, threaded and saved alongside everything else. The integration is a new front door, not a separate room — your decision history stays in one place. And every response includes a direct link back to the full Decision Board, so even when your assistant summarizes the answer in chat, you're one click from the complete synthesis, the raw model responses, and the deliberation detail.

Context that compounds

ModelCouncil is built on a simple idea: a decision tool should manage decisions, not conversations. Every query produces a structured Decision Board, and that structure is what keeps context sharp over time instead of bloating into noise. Your fiftieth query is as focused as your first.

The MCP integration extends that idea. When you query the council from Claude Desktop, you can pass along the relevant slice of your current conversation as extra grounding — and it sits alongside your project documents, not instead of them. Your assistant brings the immediate context; your ModelCouncil project brings the accumulated context. The council sees both.

That's what a generic "ask three models" feature can't replicate. The value isn't parallel model calls on their own — it's parallel model calls against context you've been building deliberately.

Getting connected

Setup takes a few minutes and depends on your client:

• Claude.ai and ChatGPT connect through their in-app Connectors interface using OAuth — paste the ModelCouncil URL, sign in, approve the connection, done. No keys to manage.
• Claude Desktop, Cursor, and Claude Code connect with a personal API key you generate under Settings → API Keys & MCP. The key is shown once at creation — copy it somewhere safe.

Full step-by-step instructions, copy-pasteable configuration for every client, and security guidance live at /docs/mcp.

Cost and control

Council queries run from MCP draw on your existing credit balance, exactly like queries run from the web app — same per-query pricing, same transparent costs. There's no separate MCP fee.

Because MCP queries run without the web app's interactive cost preview, we've built in protections: a per-call spending cap, a configurable daily cap per key, and an upfront balance check that stops a query before it runs rather than letting it complete into a negative balance. You can also tell a connector to skip a model by default — for example, "when called from Claude.ai, don't query Claude" — so you're never paying for a redundant opinion.

A beta, built carefully

The MCP server is live and labeled beta because it's new, and because the surface — multiple AI clients, OAuth flows, async deliberation — has more moving parts than the web app. We've put real work into the parts that matter most: cross-user isolation is verified end-to-end, authentication follows the OAuth 2.1 spec including PKCE and dynamic client registration, and every council query is gated and capped before it can spend a cent.

If you hit something rough, the feedback link on /docs/mcp comes straight to us. Early reports are how this gets better.

Try it

Already use ModelCouncil? Open Settings → API Keys & MCP and connect your first client. If Claude or ChatGPT is part of your daily work, this is the version of ModelCouncil that fits the way you already operate.

New here? ModelCouncil is a decision platform for people who make consequential calls — founders, executives, product and strategy leads, independent consultants. Start at modelcouncil.co, and once you're set up, the council is one connection away from the AI tools you already use.

The hard decisions don't announce themselves. Now the council is ready wherever you are when one shows up.

Read the setup guide →