add_provider
Add a new AI provider for brainstorming. Supports any OpenAI-compatible API.
LABSCO SUMMARY
The disagreement is the signal.
One model gives you one confident answer with no sense of what it is unsure about; this makes several models see and critique each other, so the synthesis surfaces the strongest disagreement rather than papering over it. The three run modes are what make it practical — hosted mode needs no keys at all, and CLI mode routes the debate through agent CLIs you already pay for, so a multi-model review does not have to mean a stack of API bills.
What it is
A server that runs multi-round debates between multiple AI models from inside your editor: 7 tools for structured debates, quick multi-model second opinions, multi-model code review, and provider management. A hosted mode needs no API keys.
What you get
- brainstorm runs a multi-round debate where models see and critique each other's answers, in API mode (your OpenAI, Gemini, DeepSeek keys), hosted mode (models already in your environment, no keys), or CLI mode through agent CLIs you already have
- brainstorm_respond lets the calling assistant submit its own substantive contribution to a round, engaging with specific points, and brainstorm_collect gathers responses in hosted mode by spawning a sub-agent per model
- brainstorm_quick fires all models in parallel and returns a compact comparison in under 10 seconds — a fast second opinion without the debate rounds or synthesis delay
- brainstorm_review is a multi-model code review: pass a diff and each model reviews independently, then findings are synthesized and deduplicated into structured results with severity, file and line references, and a verdict of approve, approve with warnings, or needs changes
- list_providers shows the configured AI providers and their default models, and add_provider adds any OpenAI-compatible API
- Debate styles beyond freeform — red-team (adversarial) and Socratic (probing questions)
Requirements
Node, run as npx brainstorm-mcp over stdio. Hosted mode needs no keys — it uses models already in your environment via sub-agents. API mode uses whichever provider keys you set, any of OPENAI_API_KEY, GEMINI_API_KEY and DEEPSEEK_API_KEY, and CLI mode runs debates through agent CLIs you already have (claude, codex) so they draw on your subscription instead of API credits.
Setup effort
One command — claude mcp add brainstorm -- npx -y brainstorm-mcp
USE CASES
01
Pressure-testing a design decision
Put the question to several models and let them debate across rounds — agreeing, disagreeing and building on each other — then read the synthesis of the recommendation, tradeoffs and where they most disagreed.
02
Getting a fast second opinion
Fire all the models in parallel for a compact comparison in seconds, for a snap decision where you just want diverse perspectives without waiting on a full debate.
03
Reviewing a pull request from several angles
Hand it a diff and have multiple models review independently, then get one deduplicated set of findings with severity and line references and a clear verdict.
04
Setting up which models take part
List the configured providers and add any OpenAI-compatible one, so the debate runs across the exact set of models you want.
7 tools
Counted once with only startup setup. Other versions, modes or optional credentials may expose a different list.
brainstorm
Run a multi-round brainstorming debate between multiple AI models. IMPORTANT: Before calling this tool, you MUST ask the user to choose a mode:
- API mode — Uses external API keys to call models (OpenAI, Gemini, DeepSeek, etc.). Best when the user has API keys configured.
- Hosted mode — No API keys needed. You execute prompts using sub-agents with models available in your environment (opus/sonnet/haiku, GPT, Gemini, etc.). Same model can be used multiple times — each run produces different perspectives.
Present these two options to the user with a one-liner explanation, then proceed based on their choice.
For API mode: set mode='api'. When participate=true (default), YOU also participate as a debater alongside external models via brainstorm_respond. For Hosted mode: set mode='hosted'. Ask the user which models to use, then spawn sub-agents for each model, collect responses, and call brainstorm_collect.
brainstorm_collect
Submit collected model responses for a hosted brainstorm session. After receiving prompts from brainstorm (mode='hosted') or a previous brainstorm_collect call, execute each prompt by spawning a sub-agent for EACH model (use the model parameter to select the right model, e.g., model='sonnet' or model='haiku'). Collect all responses and submit them here. This tool returns either: (1) the next round's prompt to execute, (2) a synthesis prompt for a single model, or (3) the final formatted debate result when complete.
brainstorm_quick
Get instant multi-model perspectives on any question — no debate rounds, no synthesis delay. Fires all models in parallel and returns a compact comparison. Under 10 seconds. Use this for quick second opinions, snap decisions, or when you want diverse perspectives fast. For deeper analysis with multiple rounds and synthesis, use the brainstorm tool instead.
brainstorm_respond
Submit YOUR (Claude's) response for the current round of an interactive brainstorm session. After the brainstorm tool returns external models' responses, call this tool with your substantive contribution. Read all responses carefully and engage with specific points — agree, disagree, build upon, or challenge ideas. Do not just summarize. After your response, the next round runs automatically (or synthesis if final round).
brainstorm_review
Multi-model code review. Pass a diff and get structured findings with severity, file/line references, and a verdict (approve / approve with warnings / needs changes). Multiple models review independently, then findings are synthesized and deduplicated. Use this for PR reviews, code audits, or pre-commit checks.
list_providers
List all configured AI providers and their default models for brainstorming.
BEFORE YOU INSTALL
What to know first
Tools you get
7 tools when we connected. With no API keys and no agent CLIs on PATH, list_providers shows an empty roster and only hosted mode works. The main tool is plain brainstorm, next to four brainstorm_* siblings.
Requires
npx on PATH
Where it runs
Both documented clients run the server on your machine.
CONTEXT COST
Light Standing cost
Rated by the size of the 7 tool definitions that ride along on every turn, not by how many there are
INSTALLATION
Connect your client
Server setup
TERMINAL
claude mcp add brainstorm -- npx -y brainstorm-mcpMCP Clients
Claude Code
CLAUDE CODE · JSON
{
"mcpServers": {
"brainstorm": {
"command": "npx",
"args": [
"-y",
"brainstorm-mcp"
],
"env": {
"OPENAI_API_KEY": "<YOUR_OPENAI_API_KEY>",
"GEMINI_API_KEY": "<YOUR_GEMINI_API_KEY>",
"DEEPSEEK_API_KEY": "<YOUR_DEEPSEEK_API_KEY>"
}
}
}
}.mcp.json
Claude Desktop
CLAUDE DESKTOP · JSON
{
"mcpServers": {
"brainstorm": {
"command": "npx",
"args": [
"-y",
"brainstorm-mcp"
],
"env": {
"OPENAI_API_KEY": "<YOUR_OPENAI_API_KEY>",
"DEEPSEEK_API_KEY": "<YOUR_DEEPSEEK_API_KEY>"
}
}
}
}
