Turn Any LLM Into Your Personal Debugging Partner

Stop explaining bugs to your AI through screenshots and code snippets. This MCP server gives Claude (or any LLM) direct control over VS Code's debugger, letting it set breakpoints, evaluate expressions, and step through your code interactively.

The Manual Debugging Loop is Broken

You know the drill: hit a bug, set some breakpoints, step through code, check variable values, form a hypothesis, test it, repeat. Then when you want AI help, you're copying stack traces and explaining context that the AI can't actually verify.

Now your AI can just debug the code directly. Set breakpoints where it thinks the issue might be, evaluate expressions to check assumptions, and iterate through the debugging process automatically.

What You Get

Interactive Debugging, Not Guesswork: Instead of Claude making educated guesses from static code, it can run your program, pause at problematic lines, and inspect actual runtime values. No more "try adding a print statement here" suggestions.

Language-Agnostic Intelligence: Works with Python, JavaScript, Go, Rust, C++, or anything else VS Code can debug. The AI doesn't need language-specific debugging knowledge—it just uses the universal debugging interface.

Expression Evaluation: The AI can evaluate arbitrary expressions in your program's context. Want to know what user.preferences[0].settings contains at runtime? The AI can check instantly without you manually typing it in the debug console.

Conditional Breakpoints: Set breakpoints that only trigger when specific conditions are met, letting the AI focus on the exact scenarios causing issues.

Real-World Debugging Scenarios

Race Conditions: Instead of adding timing logs everywhere, the AI can set breakpoints in multiple threads and track execution order to identify synchronization issues.

Data Pipeline Bugs: When processing arrays or objects fails midway through, the AI can pause at each transformation step and verify the data shape matches expectations.

API Integration Issues: The AI can inspect request/response objects at runtime, check authentication headers, and validate data transformations without you manually logging everything.

Logic Errors in Algorithms: For complex algorithms, the AI can step through iterations, check loop conditions, and validate intermediate calculations to pinpoint where logic breaks down.

Integration That Actually Works

The extension runs as both a VS Code extension and MCP server. Install it from the marketplace, configure your MCP client (Claude Desktop, Continue, Cursor), and you're ready to debug.

Your existing .vscode/launch.json configurations work unchanged. The AI gets access to the same debugging capabilities you use manually—breakpoints, step-through debugging, variable inspection, and expression evaluation.

When you hit a bug, just ask: "Debug this step by step using breakpoints and figure out where it goes wrong." The AI handles the rest, setting breakpoints, evaluating expressions, and walking through the execution flow until it identifies the root cause.

Setup for Claude Desktop:

{
  "mcpServers": {
    "debug": {
      "command": "node",
      "args": ["/path/to/mcp-debug.js"]
    }
  }
}

Setup for Continue:

{
  "experimental": {
    "modelContextProtocolServers": [{
      "transport": {
        "type": "stdio",
        "command": "node", 
        "args": ["/path/to/mcp-debug.js"]
      }
    }]
  }
}

The AI can now start debugging sessions, set breakpoints, evaluate expressions, and step through code—giving you an intelligent debugging partner that can actually see what's happening in your running program.

GitHub: jasonjmcghee/claude-debugs-for-you
VS Code Marketplace: Claude Debugs For You