Brainstorm MCP server

This is a helper that lets your AI run a small debate between several AI models instead of giving you one confident answer. You ask a question, the models answer, read each other, and argue for a round or two, and you end up with a short summary of what they agree on and where they clash. It is handy when a decision matters and you want to see the weak spots, not just one opinion.

What this MCP server does

You ask your AI to brainstorm a topic, for example whether to use one technology or another. Your AI passes the topic to this helper, and the helper sends it to the models you picked, like GPT, Gemini, DeepSeek, or a model running on your own computer. The models answer in parallel, then see each other's answers and get a chance to critique and refine them over one or more rounds. At the end, a model writes a short verdict with three parts: the recommendation, the main tradeoffs, and the strongest disagreement. You read that in the chat, along with the individual answers if you want more detail.

Click to zoom

What you can do with it

• Run a multi-round debate between several AI models on one question
• Get quick side-by-side opinions from multiple models in under 10 seconds
• Ask for a multi-model code review of a diff with a verdict and findings
• Use a red-team style where models challenge each other's answers
• Use a Socratic style where models ask probing questions instead of giving answers
• Ground the debate in your own code, diff, or architecture notes
• See which models are set up and whether their keys are working

Try asking your AI

• “Brainstorm using opus, sonnet, and haiku about whether we should use GraphQL or REST for our public API”
• “Use brainstorm_quick to compare Redis vs PostgreSQL for session storage in a Node.js app”
• “Review this diff for security issues and tell me if it needs changes before I merge”
• “Run a red-team debate about our plan to move the database to a managed service”

What it gives back to you

You get a short verdict in the chat with three bullets: the recommendation, the key tradeoffs, and the strongest disagreement between models. For code reviews you get a verdict like approve or needs changes, plus a table of findings with severity, file, line numbers, and suggestions. Quick mode gives you each model's answer side by side, with timing and which ones failed. The individual model answers are included too, so you can read the full reasoning if you want.

Before you start

Who it's for

People who make technical or design decisions and want a second, third, and fourth opinion before committing, such as developers, tech leads, and product managers.