AI Debug Bridge

Let your AI agent debug running code—not just read it.

AI Debug Bridge connects AI coding agents to the debuggers already installed in VS Code. Your agent can launch an application, set breakpoints, inspect live variables, step through execution, evaluate expressions, and read runtime logs while investigating a bug.

Get started

1. Install AI Debug Bridge and reload VS Code.
2. Open the project you want to debug in a trusted workspace.
3. Make sure the debugger for your language is installed and your project has a working launch or attach configuration.
4. Connect your preferred agent:
   • For Codex, run AI Debug Bridge: Configure Codex MCP (One Time), then start a new Codex session.
   • For Claude Code, run AI Debug Bridge: Configure Claude Code MCP (One Time), then start a new Claude Code session.
   • In GitHub Copilot Chat in VS Code, enable Debugger and Logs in the tools picker or reference #debugBridge in your prompt.
5. Tell the agent what you want to debug or launch.

Ask naturally—for example:

Debug why checkout fails when I submit an order.

Or:

Start the backend and frontend locally so I can test the app.

Keep the project and its VS Code window open while the agent works. Results are also available in the AI Debug Bridge output channel.

Debug with runtime evidence

Static code analysis can suggest what might be wrong. AI Debug Bridge helps an agent observe what is actually happening:

• Stop execution at the line that matters.
• Inspect call stacks, scopes, and live variables.
• Evaluate expressions in the current frame.
• Step over, into, and out of code.
• Continue, pause, restart, or stop a session.
• Read debugger output and application logs.

The bridge is language-agnostic and works through VS Code's standard debugger interface. Use it with Node.js, Python, Go, Java, .NET, C/C++, browser debugging, and other languages supported by an installed VS Code debug adapter.

Works with your AI coding workflow

• Codex: one-time setup from the VS Code Command Palette.
• Claude Code: one-time setup from the VS Code Command Palette.
• GitHub Copilot in VS Code: enable Debugger and Logs or reference #debugBridge.
• Other MCP clients: connect through the included local MCP bridge.
• No AI required: every debugger action is also available as a VS Code command.

Make AI Debug Bridge the preferred launcher

Add the following instruction to your project's agent guidance, such as AGENTS.md, CLAUDE.md, or another AI instruction file:

## Launching and debugging services

Whenever asked to launch or debug an application, backend, frontend, microservice, microfrontend, worker, or VS Code launch configuration, prioritize the `ai-debug-bridge` MCP over terminal-only commands so debugger controls and runtime logs remain available.

This helps your agent choose debugger-aware launches automatically instead of starting services without runtime inspection access.

Built for local development

AI Debug Bridge routes each request to the correct open workspace, including when you have multiple VS Code windows. Communication stays on the local machine, uses authenticated loopback connections, and is limited to trusted workspaces.

Your installed debug adapter remains in control of language-specific behavior. The bridge uses standard Debug Adapter Protocol operations rather than replacing your existing debugger setup.

Learn more

See the technical guide for manual MCP setup, the complete operation reference, extension APIs, configuration examples, logging behavior, security details, limitations, and contributor instructions.

Found a bug or have an idea? Open an issue.

License

MIT