SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI codingclaude-debugs-for-you

claude-debugs-for-you

VS Code extension and MCP server for language-agnostic automatic debugging via breakpoints and expression evaluation.

Open website

About this tool

Purpose

Claude Debugs For You (also known as Vibe Debugging) is designed to let Claude or other LLMs debug code interactively. It consists of an MCP server and a VS Code extension, enabling models to locate and fix issues through breakpoints and expression evaluation. The tool is language-agnostic, provided that the debugger console supports it and a valid launch.json for debugging exists in VS Code.

Key capabilities

  • Lets Claude (or other LLMs) debug interactively and evaluate expressions.
  • Works through MCP together with the VS Code extension; in theory compatible with other models and clients, but officially demonstrated only with Claude Desktop and Continue.
  • Language-agnostic, relying on debugger console support and a valid launch.json.
  • The model can identify issues on its own and suggest fixes, which the user can apply with a click.
  • Requires selecting a model with tool-calling capabilities; in the tools list, you need to click "debug" and set it to "Automatic".

How to use

  1. Download the extension from releases or the VS Code Marketplace.
  2. Install the extension. If using the .vsix directly, in VS Code's "Extensions", click the three dots and select "Install from VSIX...".
  3. After installation, a status menu item "Claude Debugs For You" will appear, showing whether it is running normally (checkmark) or failed to start (cross). Click this status menu to view available commands.
  4. Depending on your setup, choose one of the following methods:
    • In VS Code commands, search for "Copy MCP Debug Server stdio path to clipboard" to copy the stdio server path, paste it into claude_desktop_config.json (you need to update it to the actual copied path), or edit it in the same way as other MCP servers.
    • Start Claude Desktop (or another MCP client). Note: if it is already running, you may need to restart it. If you are using Continue/Cursor or another client built into VS Code, you can skip this step.
    • Use "Copy MCP Debug Server sse address to clipboard" command to get the MCP server SSE address, or you can also write it directly as http://localhost:4711/sse，或使用设置中配置的端口。
    • Add this address according to the client's requirements. Some clients (such as Cursor) may require clicking "refresh".
    • Start the MCP client. Note: if it is already running, it may need to be restarted. If you are using Continue/Cursor or another client built into VS Code, you can skip this step.
  5. Open the project containing .vscode/launch.json, and the first configuration has already been set to debug a specific file using ${file}.
  6. You can refer to "Run an Example" or watch the demo video.

Use Cases

  • Interactive debugging in VS Code with the help of an LLM for any language.
  • Development workflows that require the model to automatically locate problems and provide fix suggestions.
  • Debugging using MCP-supporting clients such as Claude Desktop, Continue, and Cursor.

Limitations and Notes

  • The language-agnostic premise is that the debugger console supports it and that a valid launch.json exists in VS Code.
  • A model with tool-calling capabilities needs to be selected.
  • Officially, only Claude Desktop and Continue are used for demonstration; other models/clients have not been explicitly verified.
  • If you see an MCP error in Continue, you can try disabling/re-enabling the Continue extension.
  • A pop-up will appear when multiple VS Code windows are started, and "Claude Debugs For You" can be gracefully handed off between windows.
  • Auto-start can be disabled, after which you need to click the status menu and select "Start Server".
  • Project todo ideas include: using ripgrep to find content instead of listing files and getting file contents; supporting conditional breakpoints; adding a "fix" tool by allowing MCP to insert CodeLens or "auto fix" suggestions, letting the user choose whether to apply the recommended changes.

Running from Source

  • Clone/open this repository using VS Code.
  • Run npm run install and npm run compile.
  • Clicking "run" will open a new VS Code.
  • The rest is the same as "Getting Started".
  • To rebuild, you can run npm run compile.
  • In the VS Code window, openexamples/python and enter the prompt.

Related tools

Browse category