VividNightmareUnleashed/claude-o3pro-mcp

Claude o3-pro MCP Integration

An MCP (Model Context Protocol) server that enables Claude Code to leverage OpenAI's o3-pro model for complex reasoning tasks.

Overview

This MCP server provides a bridge between Claude Code and OpenAI's o3-pro model, allowing you to:

• Delegate complex problems to o3-pro's deep reasoning capabilities
• Check status of long-running o3-pro requests
• Track costs with per-request pricing in cents
• Leverage Claude's codebase navigation with o3-pro's problem-solving

Key Features

• Smart Context Preparation: Claude gathers all relevant code and context, then packages it optimally for o3-pro
• Two-Phase Processing: Create request and retrieve results separately to handle long processing times
• Cost Tracking: Shows estimated and actual costs per request
• Reasoning Summaries: Optional display of o3-pro's thought process
• Environment Awareness: Automatically includes working directory, git status, and project type

Installation

1. Clone this repository:

git clone https://github.com/yourusername/claude-o3pro-mcp.git
cd claude-o3pro-mcp

2. Install dependencies:

npm install

3. Build the project:

npm run build

4. Create a .env file based on .env.example:

cp .env.example .env

5. Add your OpenAI API key to .env:

OPENAI_API_KEY=sk-your-openai-api-key-here

Configuration

Add the MCP server to Claude Code:

claude mcp add file:///path/to/claude-o3pro-mcp/dist/index.js

Or add it to your Claude Code configuration manually:

{
  "mcpServers": {
    "o3pro": {
      "command": "node",
      "args": ["/path/to/claude-o3pro-mcp/dist/index.js"],
      "env": {
        "OPENAI_API_KEY": "sk-your-key-here"
      }
    }
  }
}

Usage

Tools Available

1. @pro - Create an o3-pro request for deep analysis

   • Returns a response ID immediately
   • Claude will automatically gather relevant code and context before calling o3-pro
   • Simply describe your problem and Claude will handle the rest

2. @pro_retrieve - Check status and retrieve results

   • Use the response_id from the pro tool
   • Call repeatedly until status shows 'completed'

Best Practices

1. Let Claude gather context first: o3-pro excels at reasoning but not at navigating codebases. Let Claude collect all relevant files and context before invoking o3-pro.

2. Use for genuinely complex problems: With costs of $20/million input tokens and $80/million output tokens, reserve o3-pro for problems that truly benefit from deep reasoning.

3. Provide complete context: Since o3-pro has a 200k token context limit, ensure all necessary information is included in your query.

Example Workflow

User: I need to optimize this graph algorithm for finding shortest paths in a weighted directed graph with negative edges

Claude: I'll examine your current implementation and gather the relevant context to send to o3-pro for deep analysis.

[Claude automatically:
1. Navigates the codebase
2. Reads relevant files
3. Understands the structure
4. Calls @pro with the complete context]

🎯 o3-pro Request Created
📋 Response ID: resp_abc123...
💰 Estimated Cost: 45.0 cents
⏱️  Status: queued

💡 Next step: Use the pro_retrieve tool with response_id "resp_abc123..." to check the status and retrieve the result.

User: Check the status

Claude: [Calls @pro_retrieve with the response ID]

🧠 o3-pro is thinking deeply... Elapsed: 2m 34s

User: [After a few minutes] Check again

Claude: [Calls @pro_retrieve again]

🎯 o3-pro Analysis Complete
📋 Response ID: resp_abc123...
⏱️  Reasoning Time: 5m 23s
💰 Request Cost: 52.3 cents

💭 Response:
[Detailed solution with optimized algorithm]

How It Works

1. You describe the problem to Claude
2. Claude gathers context - navigates your codebase, reads relevant files
3. Claude calls @pro - automatically provides all gathered context to o3-pro
4. o3-pro reasons deeply - uses its full capability on the complete problem
5. You get the solution - with cost and reasoning time information

Environment Variables

• OPENAI_API_KEY (required): Your OpenAI API key
• MAX_COST_PER_SESSION: Maximum cost per session in USD (default: 10.00)
• TIMEOUT_SECONDS: Maximum time for o3-pro requests (default: 300)
• CACHE_TTL_MINUTES: Cache duration for similar queries (default: 60)

Cost Information

o3-pro pricing (as of 2025):

• Input: $20 per million tokens
• Output: $80 per million tokens (includes reasoning tokens)

Average request cost: $0.50 - $5.00 depending on problem complexity

Development

# Run in development mode
npm run dev

# Build for production
npm run build

# Start production server
npm start

Architecture

• MCP Server: Handles communication with Claude Code
• OpenAI Client: Manages o3-pro Responses API interactions
• Context Preparer: Optimizes context for o3-pro's strengths
• Progress Handler: Provides real-time status updates
• Cost Tracker: Monitors and limits spending

Troubleshooting

1. "Cost limit exceeded": Increase MAX_COST_PER_SESSION in .env
2. Timeout errors: Increase TIMEOUT_SECONDS for very complex problems
3. No progress updates: Ensure your Claude Code version supports MCP notifications

License

MIT