LLDB MCP Server

An MCP (Model Context Protocol) server that provides structured debugging tools for LLDB, designed for use with Claude Code and other MCP-compatible AI assistants.

Features

This server exposes LLDB debugging capabilities through well-defined MCP tools:

Execution Control

• lldb_run - Run a program with optional breakpoints and arguments
• lldb_analyze_crash - Analyze crash dumps and core files

Breakpoints & Watchpoints

• lldb_set_breakpoint - Set breakpoints by function, file:line, or address
• lldb_watchpoint - Set watchpoints to break on variable access

Inspection

• lldb_examine_variables - View local variables and arguments
• lldb_backtrace - Get stack traces for all threads
• lldb_registers - View CPU register values
• lldb_read_memory - Read and display memory contents
• lldb_threads - List all threads and their states

Code Analysis

• lldb_disassemble - Disassemble functions or address ranges
• lldb_source - List source code with line numbers
• lldb_symbols - Look up symbols by name, regex, or address
• lldb_images - List loaded executables and shared libraries

Expression Evaluation

• lldb_evaluate - Evaluate C/C++ expressions in debug context

Utilities

• lldb_run_command - Run arbitrary LLDB commands
• lldb_help - Get help on LLDB commands
• lldb_version - Show LLDB version info

Requirements

• Python 3.10+
• LLDB (with command-line tool in PATH)
• mcp[cli] Python package

Installing LLDB

Ubuntu/Debian:

sudo apt install lldb

macOS:

# LLDB comes with Xcode Command Line Tools
xcode-select --install

Windows:

# Install via LLVM releases or Visual Studio
winget install LLVM.LLVM

Installation

Option 1: Install from source

# Clone the repository
git clone https://github.com/yourusername/lldb-mcp.git
cd lldb-mcp

# Install dependencies
pip install -e .

Option 2: Install dependencies directly

pip install "mcp[cli]" pydantic httpx

Configuration for Claude Code

Add the following to your Claude Code MCP configuration file:

Location of config file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json
• Linux: ~/.config/Claude/claude_desktop_config.json

Configuration:

{
  "mcpServers": {
    "lldb": {
      "command": "python",
      "args": ["/path/to/lldb-mcp/lldb_mcp_server.py"]
    }
  }
}

Or if installed as a package:

{
  "mcpServers": {
    "lldb": {
      "command": "lldb-mcp"
    }
  }
}

Using uvx (recommended for isolation):

{
  "mcpServers": {
    "lldb": {
      "command": "uvx",
      "args": ["--from", "/path/to/lldb-mcp", "lldb-mcp"]
    }
  }
}

Usage Examples

Once configured, you can ask Claude Code to help with debugging tasks:

Analyze a Crash

"Analyze the crash dump in ./core and the executable ./myprogram to find what caused the segfault"

Set Breakpoints and Examine State

"Set a breakpoint at the processData function in processor.cpp, run the program with argument 'test.txt', and show me the local variables when it stops"

Disassemble Code

"Show me the assembly for the main function in ./myprogram"

Evaluate Expressions

"Run ./myprogram until it hits parseConfig and evaluate the expression config->max_threads"

Memory Inspection

"Read 128 bytes of memory at address 0x7fff5fbff000 in hexadecimal format"

Symbol Lookup

"Find all symbols matching 'parse.*' regex in ./myprogram"

Tool Reference

lldb_run_command

Execute any LLDB command directly.

{
    "command": "help breakpoint",  # Any LLDB command
    "target": "./myprogram",       # Optional: executable to load
    "working_dir": "/path/to/dir"  # Optional: working directory
}

lldb_analyze_crash

Analyze crash dumps with full context.

{
    "executable": "./myprogram",
    "core_file": "./core",              # Optional: core dump
    "response_format": "markdown"       # or "json"
}

lldb_set_breakpoint

Set breakpoints with conditions.

{
    "executable": "./myprogram",
    "location": "main.cpp:42",          # or "functionName" or "0x400500"
    "condition": "i > 100"              # Optional: break condition
}

lldb_examine_variables

View variables at a breakpoint.

{
    "executable": "./myprogram",
    "breakpoint": "processData",
    "variables": ["buffer", "size"],    # Optional: specific vars
    "args": ["input.txt"],              # Optional: program args
    "response_format": "markdown"
}

lldb_disassemble

Disassemble code regions.

{
    "executable": "./myprogram",
    "target": "main",                   # Function name, address range, or "current"
    "show_bytes": true,                 # Show opcode bytes
    "mixed": true                       # Interleave source
}

lldb_read_memory

Read memory contents.

{
    "executable": "./myprogram",
    "address": "0x7fff5fbff000",
    "count": 64,                        # Bytes to read
    "format": "x",                      # x=hex, b=binary, d=decimal, s=string
    "breakpoint": "main"                # Optional: stop here first
}

lldb_evaluate

Evaluate C/C++ expressions.

{
    "executable": "./myprogram",
    "expression": "ptr->data[5]",
    "breakpoint": "processBuffer",
    "args": ["test.dat"]
}

lldb_backtrace

Get stack traces.

{
    "executable": "./myprogram",
    "breakpoint": "handleError",        # or use core_file
    "core_file": "./core",              # For post-mortem
    "all_threads": true,
    "limit": 50,
    "response_format": "json"           # Structured output
}

lldb_registers

View CPU registers.

{
    "executable": "./myprogram",
    "breakpoint": "criticalSection",
    "register_set": "general",          # general, float, vector, all
    "specific_registers": ["rax", "rbx", "rsp"]  # Optional
}

lldb_watchpoint

Set data watchpoints.

{
    "executable": "./myprogram",
    "variable": "global_counter",
    "watch_type": "write",              # write, read, read_write
    "condition": "global_counter > 1000"
}

lldb_symbols

Look up symbols.

{
    "executable": "./myprogram",
    "query": "process.*",
    "query_type": "regex"               # name, regex, address, type
}

Alternative: Using LLDB's Built-in MCP Server

LLDB 18+ has built-in MCP support. To use it instead:

1. Start LLDB and enable MCP:

   (lldb) protocol-server start MCP listen://localhost:59999
   

2. Configure Claude Code to connect via netcat:

   {
     "mcpServers": {
       "lldb": {
         "command": "/usr/bin/nc",
         "args": ["localhost", "59999"]
       }
     }
   }
   

Note: LLDB's built-in MCP only exposes a single lldb_command tool, whereas this server provides structured, specialized tools for better AI integration.

Development

Running Tests

pytest tests/

Type Checking

mypy lldb_mcp_server.py

Linting

ruff check lldb_mcp_server.py
ruff format lldb_mcp_server.py

Troubleshooting

"LLDB executable not found"

Ensure LLDB is installed and in your PATH:

which lldb
lldb --version

Permission denied on core files

On Linux, enable core dumps:

ulimit -c unlimited
sudo sysctl -w kernel.core_pattern=core.%p

Debugger can't find symbols

Compile your programs with debug info:

g++ -g -O0 myprogram.cpp -o myprogram
clang++ -g -O0 myprogram.cpp -o myprogram

License

MIT License - see LICENSE file for details.

Contributing

Contributions welcome! Please read CONTRIBUTING.md for guidelines.