Claude Debugs for You
Enable Claude (or any other LLM) to interactively debug your code (set breakpoints and evaluate expressions in stack frame). It's language-agnostic, assuming debugger console support and valid launch.json for debugging in VSCode.
jasonjmcghee
README
<img src="./images/claude-debugs-for-you.png" width="64" height="64" alt="description" align="center"> Claude Debugs For You
aka Vibe Debugging
Enable Claude (or any other LLM) to interactively debug your code
This is an MCP Server and VS Code extension which enables claude to interactively debug and evaluate expressions.
That means it should also work with other models / clients etc. but I only demonstrate it with Claude Desktop and Continue here.
It's language-agnostic, assuming debugger console support and valid launch.json for debugging in VSCode.
Getting Started
- Download the extension from releases or VS Code Marketplace
- Install the extension
- If using
.vsix
directly, go to the three dots in "Extensions" in VS Code and choose "Install from VSIX..."
- You will see a new status menu item "Claude Debugs For You" which shows if it is running properly (check) or failed to startup (x)
<img width="314" alt="Screenshot 2025-03-22 at 9 51 22 PM" src="https://github.com/user-attachments/assets/2cd65e0d-4c1d-4fb6-b9ea-3995149b4043" />
You can click this status menu for the commands available
<img width="510" alt="Screenshot 2025-03-22 at 9 59 22 PM" src="https://github.com/user-attachments/assets/54e339e3-81f8-4ef2-a201-6742aa2c97a8" />
Follow one of the options below, depending on your setup
<details> <summary>If using stdio (classic, required for Claude Desktop)</summary>
-
Copy the stdio server path to your clipboard by searching vs code commands for "Copy MCP Debug Server stdio path to clipboard"
-
Paste the following (BUT UPDATE THE PATH TO THE COPIED ONE!) in your
claude_desktop_config.json
or edit accordingly if you use other MCP servers
{
"mcpServers": {
"debug": {
"command": "node",
"args": [
"/path/to/mcp-debug.js"
]
}
}
}
- Start Claude desktop (or other MCP client)
- Note: You may need to restart it, if it was already running.
- You can skip this step if using Continue/Cursor or other built-in to VS Code </details>
<details>
<summary>If using /sse
(e.g. Cursor)</summary>
- Retrieve the MCP server sse address by using the "Copy MCP Debug Server sse address to clipboard" command
- You can just write it out server URL of "http://localhost:4711/sse", or whatever port you setup in settings.
- Add it wherever you need to based on your client
- You may need to hit "refresh" depending on client: this is required in Cursor
- Start MCP client
- Note: You may need to restart it, if it was already running.
- You can skip this step if using Continue/Cursor or other built-in to VS Code
</details>
You're ready to debug!
VS Code Debugging Documentation
Open a project containing a .vscode/launch.json
with the first configuration setup to debug a specific file with ${file}
.
See Run an Example below, and/or watch a demo video.
Contributing
Find bugs or have an idea that will improve this? Please open a pull request or log an issue.
Does this readme suck? Help me improve it!
Demo
Using Continue
It figures out the problem, and then suggests a fix, which we just click to apply
https://github.com/user-attachments/assets/3a0a879d-2db7-4a3f-ab43-796c22a0f1ef
<details> <summary>How do I set this up with Continue? / Show MCP Configuration</summary>
Configuration:
{
...
"experimental": {
"modelContextProtocolServers": [
{
"transport": {
"type": "stdio",
"command": "node",
"args": [
"/Users/jason/Library/Application Support/Code/User/globalStorage/jasonmcghee.claude-debugs-for-you/mcp-debug.js"
]
}
}
]
}
}
You'll also need to choose a model capable of using tools.
When the list of tools pops up, make sure to click "debug" in the list of your tools, and set it to be "Automatic".
Troubleshooting
If you are seeing MCP errors in continue, try disabling / re-enabling the continue plugin
</details>
If helpful, this is what my configuration looks like! But it's nearly the same as Claude Desktop.
Using Claude Desktop
In this example, I made it intentionally very cautious (make no assumptions etc - same prompt as below) but you can ask it to do whatever.
https://github.com/user-attachments/assets/ef6085f7-11a2-4eea-bb60-b5a54873b5d5
Developing
- Clone / Open this repo with VS Code
- Run
npm run install
andnpm run compile
- Hit "run" which will open a new VSCode
- Otherwise same as "Getting Started applies"
- To rebuild,
npm run compile
Package
vsce package
Run an Example
Open examples/python
in a VS Code window
Enter the prompt:
i am building `longest_substring_with_k_distinct` and for some reason it's not working quite right. can you debug it step by step using breakpoints and evaluating expressions to figure out where it goes wrong? make sure to use the debug tool to get access and debug! don't make any guesses as to the problem up front. DEBUG!
Other things worth mentioning
When you start multiple vs code windows, you'll see a pop-up. You can gracefully hand-off "Claude Debugs For You" between windows.
You can also disable autostart. Then you'll just need to click the status menu and select "Start Server".
<img width="395" alt="Screenshot 2025-03-22 at 10 08 52 PM" src="https://github.com/user-attachments/assets/2b6d1b61-a2c6-4447-8054-b4dd02a716e8" />
Short list of ideas
- [ ] It should use ripgrep to find what you ask for, rather than list files + get file content.
- [x] Add support for conditional breakpoints
- [ ] Add "fix" tool by allowing MCP to insert a CodeLens or "auto fix" suggestion so the user can choose to apply a recommended change or not.
- Your idea here!
Recommended Servers

VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
AIO-MCP Server
🚀 All-in-one MCP server with AI search, RAG, and multi-service integrations (GitLab/Jira/Confluence/YouTube) for AI-enhanced development workflows. Folk from
Hyperbrowser MCP Server
Welcome to Hyperbrowser, the Internet for AI. Hyperbrowser is the next-generation platform empowering AI agents and enabling effortless, scalable browser automation. Built specifically for AI developers, it eliminates the headaches of local infrastructure and performance bottlenecks, allowing you to
BigQuery
This is a server that lets your LLMs (like Claude) talk directly to your BigQuery data! Think of it as a friendly translator that sits between your AI assistant and your database, making sure they can chat securely and efficiently.
Perplexity Chat MCP Server
MCP Server for the Perplexity API.
Web Research Server
A Model Context Protocol server that enables Claude to perform web research by integrating Google search, extracting webpage content, and capturing screenshots.
MySQL Server
Allows AI assistants to list tables, read data, and execute SQL queries through a controlled interface, making database exploration and analysis safer and more structured.
Aindreyway Codex Keeper
Serves as a guardian of development knowledge, providing AI assistants with curated access to latest documentation and best practices.
Etherscan Tools
Facilitates interaction with Ethereum blockchain data via Etherscan's API, providing real-time access to balances, transactions, token transfers, contract ABIs, gas prices, and ENS name resolutions.
Perplexity Deep Research
A server that allows AI assistants to perform web searches using Perplexity's sonar-deep-research model with citation support.