mcp-debugger

A debugger with no IDE: an agent sets breakpoints, steps through code and inspects variables in ten languages

  • 188 ★ on the repository
  • Last commit
  • MIT
MCP server

Broad access

Why this level

  • Debugging runs your program with the agent rights and evaluated expressions execute code inside it
  • Attaching to processes on other machines and pods exposes their state and data
All reasons and checks

debugmcp/mcp-debugger

Install

Manual install

npx -y @debugmcp/mcp-debugger stdio

Runs over stdio; clients start it themselves.

Access and checks

How we review

Broad accessThe tool can change data, run commands or handle secrets.

  • Debugging runs your program with the agent rights and evaluated expressions execute code inside it
  • Attaching to processes on other machines and pods exposes their state and data

This is third-party code. Review the repository files before installing.

What it does

mcp-debugger turns step-through debugging into MCP calls. The agent runs a program under the real language debugger through the Debug Adapter Protocol, sets breakpoints, steps, evaluates expressions and inspects the stack and variables. Supported are Python (debugpy), JavaScript and Node.js (js-debug), Ruby (rdbg), Rust and C/C++ (CodeLLDB), Go (Delve), Java (JDI), .NET (netcoredbg), COBOL and Dart with Flutter, including hot reload. No IDE is needed: the server works anywhere Node.js runs, that is CI, containers, Kubernetes pods, SSH boxes and cloud agent sandboxes. It can attach to a process on another machine without an IDE and has logpoints that log without pausing the program. Transport is stdio or Streamable HTTP and every debug session gets its own process. By default variable, expression and output values are masked for secrets. The repository installs as an npm package, a Docker image and a skill.

Who it is for. Developers whose agent works in a terminal, CI or cloud sandbox and must find bugs with a debugger and not by adding prints.

Good fit when

  • A test fails only in a container and the agent attaches a debugger to a process in a Kubernetes pod
  • An exception is rare and the agent sets a conditional breakpoint and inspects local variables
  • A Flutter app behaves oddly and the agent changes code with hot reload
  • Production cannot be paused so logpoints record values to a log
  • Legacy COBOL is stepped through and WORKING-STORAGE values are inspected

Not a fit when

  • The agent works inside an open VS Code: a server that drives the editor debugger fits better
  • You need a PHP debugger: that language is not in the list
  • The bug is visible from a log and trace: launching a debugger is overkill

Example request

Run `app.py` under the debugger, set a breakpoint in the `parse_order` function and show variable values at the failure

MCP

Transport
stdio, http
Authentication
not required

Limitations

The server launches the language debugger, which must be available on the machine: Java needs JDK 21 or newer, Rust on Windows needs the GNU toolchain and C/C++ goes through CodeLLDB. Debugging runs your program with the agent rights and `evaluate` expressions execute code inside it. Secret masking is on by default but make sure parameters do not turn it off. PHP is not supported. Some features, such as Java hot swap and Flutter reload, depend on the language adapter.

How to disable. Remove the server in the client (`claude mcp remove mcp-debugger`) and stop any running debug sessions.

FAQ

How do I connect mcp-debugger?

Add the command `npx -y @debugmcp/mcp-debugger stdio` to the client MCP config. For Codex: `codex mcp add mcp-debugger -- npx -y @debugmcp/mcp-debugger stdio`.

Which languages are supported?

Python, JavaScript and TypeScript, Ruby, Rust, Go, Java, .NET, C and C++, COBOL, Dart and Flutter.

Is VS Code required?

No, the server runs as a standalone Node.js process. An `expose_session` mode shows the agent session in an IDE as read-only.

Can I debug a process in Kubernetes?

Yes. Pods connect through port-forward for debugpy, rdbg and JDWP, and native code has a `kubectl debug` flow with an ephemeral container.

Do secrets leak into agent answers?

By default variable, expression and output values are masked and a least-privilege mode restricts access further. Check the settings before running on live data.

Discussion

No questions or reviews yet.

A debugger MCP server: the agent sets breakpoints, steps through code and reads variables of a live program over DAP.

MCP serverBroad access107Repository stars

An MCP server for the Node.js debugger: breakpoints, stepping, variable inspection and expression evaluation at a pause

MCP serverBroad access350Repository stars
Official

The official Microsoft MCP server for debugging: a VS Code extension and a CLI give an agent breakpoints, steps, variables and expression evaluation

MCP serverBroad access581Repository stars

A JetBrains IDE plugin that hands the debugger to an agent: breakpoints, stepping, variables and expression evaluation over MCP.

MCP serverBroad access118Repository stars

A VS Code extension with an MCP server: an agent sets breakpoints, starts debugging from your launch.json and evaluates expressions in any language

MCP serverBroad access515Repository stars

The dbg debugger CLI for AI agents: breakpoints, steps and variables through short @ref handles instead of print debugging

CLIExtended access159Repository stars

README in short

The README compares the server with a VS Code based debugger in a feature table, lists supported languages and leads to installation through npx, Docker and a skill. Separate sections cover Kubernetes, Dart and Flutter, logpoints and secret masking.

README badge

Report a problem