re-mcp-ghidra 3.0.0

Ghidra backend for re-mcp — headless Ghidra MCP server using pyhidra

re-mcp-ghidra

Ghidra backend for RE-MCP — a headless Ghidra MCP server using pyhidra. Exposes Ghidra's analysis capabilities over the Model Context Protocol, letting LLMs drive reverse engineering directly.

This is a standalone server, not a Ghidra plugin. It uses pyhidra to run Ghidra's analysis engine without a GUI.

Requirements

• Python 3.12+
• Ghidra 11+
• JDK 21+
• macOS, Windows, or Linux

Installation

uv tool install re-mcp-ghidra

Or with pip:

pip install re-mcp-ghidra

Finding Ghidra

The server looks for your Ghidra installation in the following order:

1. GHIDRA_INSTALL_DIR environment variable — set this if Ghidra is in a non-standard location.
2. Config file — ghidra-install-dir in ~/.ghidra/ghidra-config.json.
3. Platform-specific default paths (e.g. /Applications/ghidra_* on macOS).

See the main documentation for the full list of default search paths per platform.

Usage

# Run the server (direct stdio mode)
re-mcp-ghidra

# Or with uvx (no install needed)
uvx re-mcp-ghidra

MCP client configuration

{
  "mcpServers": {
    "ghidra": {
      "command": "uvx",
      "args": ["re-mcp-ghidra"]
    }
  }
}

CLI subcommands

Command	Description
re-mcp-ghidra (or re-mcp-ghidra stdio)	Direct stdio mode — single-session, workers die on disconnect (default)
re-mcp-ghidra proxy	Stdio proxy that auto-spawns a persistent HTTP daemon
re-mcp-ghidra serve	Start the HTTP daemon directly
re-mcp-ghidra stop	Gracefully shut down a running daemon

Environment variables

Variable	Default	Description
GHIDRA_INSTALL_DIR	(auto-detected)	Path to Ghidra installation directory
GHIDRA_MCP_MAX_WORKERS	(unlimited)	Maximum simultaneous databases (1-8)
GHIDRA_MCP_LOG_LEVEL	WARNING	Logging level
GHIDRA_MCP_LOG_DIR	(unset)	Directory for per-run log files
GHIDRA_MCP_IDLE_TIMEOUT	300	Auto-shutdown timeout in seconds (0 to disable)

Features

• Full decompilation and disassembly
• Function, type, and structure management
• Cross-reference and call graph analysis
• String and byte pattern search
• Binary patching and instruction assembly
• Function ID and data type archive support
• Multi-database support with concurrent analysis
• MCP resources for structured read-only access

See the main documentation for the full tool catalog, multi-database workflows, and detailed usage.

License

Dual-licensed under MIT and Apache-2.0.