BigBrain MCP Server

When your AI needs a bigger brain.

Overview

BigBrain is an MCP server that gives your AI an elegant exit when it gets stuck, loops, or loses context. It intelligently extracts relevant code and context, packages it cleanly, and operates in three modes:

• External AI Mode: Get fresh insights from external AI models like ChatGPT 5 Pro, Grok Heavy, or Claude Opus - then return with the solution to your original chat
• Multi-Agent Mode: Enable seamless AI-to-AI communication within systems like Claude Code for direct agent collaboration
• ChatGPT Desktop Mode (Experimental): Fully automated Agent-to-Agent communication with ChatGPT Desktop app (macOS only)

Features

• Intelligent Context Extraction: Uses Probe to automatically discover and extract relevant code, dependencies, and context
• Three Operation Modes:
  • External AI Mode: Round-trip workflow with clipboard integration for user-driven external consultation
  • Multi-Agent Mode: Direct AI-to-AI communication for automated agent collaboration
  • ChatGPT Desktop Mode: Native integration with ChatGPT Desktop for Agent-to-Agent workflows (macOS)
• Clean Context Packaging: Formats everything into minimal, structured prompt packs
• Multi-Model Support: Works with any AI but optimized for advanced models like ChatGPT 5 Pro, Grok Heavy, and Claude Opus
• File-Based Communication: Reliable agent-to-agent transfer in multi-agent systems
• Configurable Notifications: Optional sound and desktop notifications (can be disabled with flags)

Installation

Claude Code (Recommended)

claude mcp add -- npx -y @probelabs/big-brain@latest

Cursor & Other MCP Clients

Add to your MCP configuration file:

{
  "mcpServers": {
    "big-brain": {
      "command": "npx",
      "args": ["-y", "@probelabs/big-brain@latest"]
    }
  }
}

Usage Modes

BigBrain operates in three distinct modes depending on your workflow needs:

External AI Mode (Default)

For manual consultation with external AI models:

# Standard installation uses External AI Mode
claude mcp add -- npx -y @probelabs/big-brain@latest

Workflow:

1. Ask your AI to use BigBrain when stuck
2. BigBrain extracts code and copies to clipboard
3. Paste into external AI (ChatGPT 5 Pro, Grok Heavy, Claude Opus)
4. Return with solution to original chat

Multi-Agent Mode

For automated AI-to-AI communication in systems like Claude Code:

# Enable loop mode with custom agent prompt
npx @probelabs/big-brain --loop "Now call research-agent to investigate this issue"

Workflow:

1. AI uses BigBrain with --loop flag and custom prompt
2. BigBrain extracts code and saves to file (no clipboard)
3. Returns instructions for the next agent to read the file
4. Seamless automated agent collaboration

ChatGPT Desktop Mode (Experimental)

For fully automated Agent-to-Agent communication with ChatGPT Desktop:

# Enable ChatGPT Desktop integration
claude mcp add -- npx -y @probelabs/big-brain@latest --chatgpt

# Use with specific terminal (if Terminal.app doesn't have permissions)
claude mcp add -- npx -y @probelabs/big-brain@latest --chatgpt --terminal iTerm

Requirements:

• macOS 11.0 or later
• ChatGPT Desktop app installed
• Terminal accessibility permissions enabled (System Settings → Privacy & Security → Accessibility)

Workflow:

1. Your AI agent calls BigBrain when it needs help
2. BigBrain automatically opens ChatGPT Desktop app
3. Sends the query and waits for response (30 seconds to 20 minutes for ChatGPT Pro thinking)
4. Returns ChatGPT's response back to your agent
5. Fully automated - no manual intervention required

Optimized for ChatGPT 5 Pro: This mode is specifically designed to leverage ChatGPT 5 Pro's advanced thinking capabilities for complex reasoning tasks.

When to Use Each Mode

Scenario	Mode	Why
Stuck on complex debugging	External AI	Get fresh human-guided analysis from advanced models
Need specialized expertise	External AI	Leverage specific model strengths (GPT-5 Pro, Grok Heavy, etc.)
Multi-agent workflows	Multi-Agent	Automated collaboration between AI agents
Claude Code environments	Multi-Agent	Direct agent-to-agent handoffs
Need ChatGPT 5 Pro thinking	ChatGPT Desktop	Automated access to ChatGPT's advanced reasoning
macOS with ChatGPT Desktop	ChatGPT Desktop	Native integration for seamless Agent-to-Agent flow
Manual consultation	External AI	Human-in-the-loop decision making
Automated pipelines	Multi-Agent	No human intervention required

Quick Start

External AI Mode

1. Ask AI to use BigBrain

When your AI gets stuck, loops, or loses context:

Ask BigBrain to help me with this React component issue

2. Get External Advice

BigBrain extracts code and copies everything to clipboard. Open fresh AI chat, paste, and get the solution.

3. Return with Solution

Copy the external AI's response and return to your original chat. Paste the solution to continue with fresh insights.

Multi-Agent Mode

1. AI Initiates with Loop Flag

Your AI system uses BigBrain with a custom agent prompt:

AI calls: bigbrain --loop "Now call research-agent to investigate this issue"

2. Direct Agent Communication

BigBrain responds with file path and instructions for the next agent to read and process the context.

3. Automated Continuation

The designated agent reads the file and continues the workflow without human intervention.

Configuration Options

Available Flags

BigBrain supports several command-line flags to customize behavior:

• --chatgpt - Enable ChatGPT Desktop automation mode
• --terminal <app> - Specify terminal for ChatGPT mode (Terminal, iTerm, Warp, Alacritty)
• --loop <prompt> - Enable loop mode with custom agent prompt
• --disable-sound - Disable sound notifications
• --disable-notification - Disable system notifications
• --max-tokens <number> - Set maximum token limit (default: 40000)

Example configurations:

{
  "mcpServers": {
    "big-brain": {
      "command": "npx",
      "args": [
        "-y", "@probelabs/big-brain@latest",
        "--disable-sound",
        "--disable-notification"
      ]
    }
  }
}

Flag Reference:

• --chatgpt: Enable ChatGPT Desktop mode (macOS only, experimental)
• --loop "custom prompt": Enable multi-agent mode with custom agent instruction
• --disable-sound: Disable sound notifications
• --disable-notification: Disable desktop notifications

Multi-Agent Configuration Example

For systems like Claude Code that need automated agent communication:

{
  "mcpServers": {
    "big-brain": {
      "command": "npx", 
      "args": [
        "-y", "@probelabs/big-brain@latest",
        "--loop", "Now call research-agent to investigate this issue",
        "--disable-sound",
        "--disable-notification"
      ]
    }
  }
}

ChatGPT Desktop Configuration Example

For macOS users with ChatGPT Desktop app installed:

{
  "mcpServers": {
    "big-brain": {
      "command": "npx",
      "args": [
        "-y", "@probelabs/big-brain@latest",
        "--chatgpt",
        "--disable-sound"
      ]
    }
  }
}

How BigBrain Works

External AI Mode Process

1. AI Preparation: Your AI prepares a comprehensive question with all needed context and details
2. Intelligent Extraction: BigBrain uses Probe to automatically scan and extract relevant code, dependencies, and context
3. Clean Packaging: Everything is formatted into a minimal, paste-ready prompt pack and copied to your clipboard
4. External Consultation: You paste the context pack into a fresh AI chat (ChatGPT 5 Pro, Grok Heavy, Claude Opus, etc.) to get analysis and solutions
5. Return Journey: Copy the external AI's response and paste it back into your original chat to continue with fresh insights

Multi-Agent Mode Process

1. Agent Initiation: AI system calls BigBrain with --loop flag and custom agent prompt
2. Intelligent Extraction: BigBrain uses Probe to automatically extract relevant code and context
3. File-Based Packaging: Context is saved to a temporary file in structured XML format
4. Agent Instruction: BigBrain responds with the custom prompt and file path for the next agent
5. Direct Handoff: The designated agent reads the file and continues processing without human intervention

ChatGPT Desktop Mode Process

1. Agent Request: Your AI agent calls BigBrain when it encounters a complex problem
2. Intelligent Extraction: BigBrain uses Probe to extract relevant code and context
3. Automated UI Control: BigBrain opens ChatGPT Desktop app via macOS accessibility APIs
4. Query Submission: Automatically types or pastes the query into ChatGPT
5. Response Capture: Waits for and captures ChatGPT's response (optimized for thinking models)
6. Agent Return: Returns the response directly to your AI agent

Why All Modes Work

External AI Mode Benefits:

• Breaks Context Loops: Fresh AI session without chat history noise
• Leverages Advanced Models: Use the best AI for complex analysis
• Human Oversight: Manual control over consultation process
• Model Flexibility: Choose the right AI for each specific problem

Multi-Agent Mode Benefits:

• Automated Workflows: No human intervention required
• Reliable Transfer: File-based communication eliminates clipboard dependencies
• Agent Coordination: Custom prompts guide specific agent actions
• Seamless Integration: Built for systems like Claude Code with multiple AI agents

ChatGPT Desktop Mode Benefits:

• Native Integration: Direct control of ChatGPT Desktop app
• Thinking Models: Optimized for ChatGPT 5 Pro's advanced reasoning
• Fully Automated: Agent-to-Agent without human intervention
• macOS Native: Leverages system accessibility for seamless control

Shared Benefits:

• Intelligent Context: Probe ensures all relevant code is included automatically
• Clean Separation: Focused context without noise
• Maintains Workflow: Continue original work with fresh insights

Optional: Automate BigBrain Usage

Add this to your CLAUDE.md file so your AI automatically knows when to use BigBrain:

# BigBrain Integration
Whenever you get stuck and can't find a solution - just ask BigBrain MCP for advice.

Development

Local Development Setup

git clone https://github.com/probelabs/big-brain.git
cd big-brain
npm install
npm run build

Development Commands

npm run build        # Build the project
npm run watch        # Watch mode for development
npm run inspector    # Test with MCP Inspector

The MCP Inspector provides a browser-based interface for testing and debugging the server.

Requirements

• Node.js 16 or higher
• npm or yarn

Technical Details

Built on Probe

BigBrain leverages Probe for intelligent code extraction. When your AI mentions specific files or functions, Probe automatically discovers and includes all related dependencies, types, and context - no manual file specification required.

Supported File Types

BigBrain works with all common programming languages and file types:

• TypeScript/JavaScript (.ts, .tsx, .js, .jsx)
• Python (.py)
• Rust (.rs)
• Go (.go)
• JSON (.json)
• Markdown (.md)
• And many more...

Troubleshooting & FAQ

Permission Errors with ChatGPT Desktop Mode

If you see osascript permission errors (-1743, -25211, or "Not authorized to send Apple events"):

1. Grant Terminal Accessibility Permissions:

   • Go to System Settings → Privacy & Security → Accessibility
   • Add and enable your terminal app (Terminal.app, iTerm, etc.)
   • Restart your terminal after granting permissions

2. Try a Different Terminal:

   • If Terminal.app doesn't work, try iTerm with --terminal iTerm
   • Example: claude mcp add -- npx -y @probelabs/big-brain@latest --chatgpt --terminal iTerm
   • Supported terminals: Terminal, iTerm, Warp, Alacritty

3. Reset Permissions (if still having issues):

   tccutil reset AppleEvents
   tccutil reset Accessibility

   Then re-grant permissions in System Settings

4. Check Terminal Has Full Disk Access:

   • System Settings → Privacy & Security → Full Disk Access
   • Enable your terminal app if needed

ChatGPT Desktop Not Responding

• Ensure ChatGPT Desktop app is installed and running
• Try closing and reopening ChatGPT Desktop
• Check that you're signed in to ChatGPT

General Connection Issues

• Verify BigBrain is properly installed: npm list @probelabs/big-brain
• Check MCP server status: claude mcp list
• Restart Claude Code if the MCP connection is lost

License

MIT

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Support

For issues and questions, please use the GitHub issue tracker.