Lldb MCP

🚀 lldb_mcp

A feature - rich and minimally - dependent simple lldb mcp server with concise code and comprehensive functionality! It only includes two commands, lldb_init and lldb, without cluttering your toolset. It works excellently in conjunction with o4 - mini and Gemini 2.5 Pro. You always maintain full control over the process.

Output is automatically captured, eliminating the need for manual copy - pasting.

🚀 Quick Start

📦 Installation

git clone https://github.com/stableversion/lldb_mcp

cd lldb_mcp && python3 -m venv venv && source venv/bin/activate && pip3 install fastmcp

Configuration Example

{
  "mcpServers": {
    "lldb": {
      "command": "/path/to/lldb_mcp/venv/bin/python",
      "args": ["/path/to/lldb_mcp/lldb_mcp.py"]
    }
  }
}

Import Command Script in LLDB

(lldb) command script import /path/to/lldb_mcp/lldb_server.py

🔧 Technical Details

System Architecture

The entire codebase is less than 200 lines, for the following reasons:

• Synchronous Debugger: Calling debugger.SetAsync(False) causes HandleCommand to block and directly capture all output. This avoids complex event listeners (SBListener) and polling.
• Simple Controller: LLDBController executes commands via HandleCommand and returns the full received output/error.
• Standard Library: It uses Python's built - in socketserver and json libraries.

Security Considerations

Since lldb can execute arbitrary commands, restrictions are necessary:

• You must manually start the server in the lldb session.
• Some commands are blacklisted. Check the lldb_server.py file.
• Command chaining is not allowed (o4 - mini particularly likes to do this).

📄 Notes

• o4 mini is prone to having --address illusions in memory commands.
• Some commands "succeed" without output, which can confuse the LLM. Therefore, "Execution successful" will be returned.
• "Continue" may run indefinitely until a breakpoint is encountered, meaning the mcp may wait indefinitely for output (using skip and continue can be helpful).