Skip to main content

Getting Started

Get debug-mcp running and perform your first debugging session in minutes.

See It in Action​

Loading terminal recording...

Prerequisites​

Install​

The recommended way to run debug-mcp is via dnx:

dnx -y debug-mcp

Configure Your AI Agent​

Claude Code / Claude Desktop​

Add to your MCP configuration (settings.json or claude_desktop_config.json):

{
  "mcpServers": {
    "dotnet-debugger": {
      "command": "dnx",
      "args": ["-y", "debug-mcp"]
    }
  }
}

Other MCP Clients​

debug-mcp communicates via MCP over stdio. Any MCP-compatible client can use it:

dnx -y debug-mcp

Verify It Works​

Ask your AI agent:

"What debug-mcp tools are available?"

You should see a list of 34 tools including debug_launch, breakpoint_set, and evaluate.

Your First Debugging Session​

Let's debug a simple .NET application step by step.

1. Launch the application​

"Launch /path/to/MyApp.dll and stop at the entry point"

The process starts and pauses immediately at Main().

Tool call details

Request (debug_launch):

{
  "program": "/path/to/MyApp.dll",
  "stop_at_entry": true
}

2. Set a breakpoint​

"Set a breakpoint at line 25 in Program.cs"

The agent maps the file and line to an IL offset and creates the breakpoint.

Tool call details

Request (breakpoint_set):

{
  "file": "Program.cs",
  "line": 25
}

3. Continue and wait​

"Continue execution and wait for the breakpoint"

The agent calls debug_continue, then breakpoint_wait. Execution resumes until line 25 is hit.

4. Inspect variables​

"Show me the local variables"

The agent retrieves all local variables, arguments, and this for the current frame.

Tool call details

Request (variables_get):

{
  "scope": "all"
}

5. Evaluate an expression​

"What is the value of items.Count?"

The agent evaluates the expression in the context of the stopped thread and returns the result.

Tool call details

Request (evaluate):

{
  "expression": "items.Count"
}

6. Disconnect​

"Stop debugging"

The agent terminates the process and ends the session.

Tool call details

Request (debug_disconnect):

{
  "terminate": true
}

Static Code Analysis​

debug-mcp also provides Roslyn-based code analysis tools that work without running the debugger.

Load a solution​

"Load the solution at /path/to/MyApp.sln for code analysis"

Tool call details

Request (code_load):

{
  "path": "/path/to/MyApp.sln"
}

Find all usages​

"Find all usages of the UserService class"

Tool call details

Request (code_find_usages):

{
  "name": "MyApp.Services.UserService",
  "symbolKind": "Type"
}

Check for errors​

"Are there any compilation errors?"

Tool call details

Request (code_get_diagnostics):

{
  "minSeverity": "Error"
}

What's Next?​