Ghidra MCP
Open-source MCP server bridging Ghidra's reverse engineering engine with AI assistants for binary decompilation and analysis
Overview
Ghidra MCP is an open-source Model Context Protocol server that connects NSA's Ghidra reverse engineering suite to AI assistants and automation frameworks. A Java plugin exposes Ghidra's analysis engine over HTTP, while a Python MCP bridge translates those endpoints into standardized MCP calls, letting tools like Claude decompile functions, trace call graphs, map cross-references, and rename or type variables directly inside a live Ghidra project. It ships both a GUI plugin for interactive use and a headless server for CI pipelines, with Docker support and lazy tool loading to keep context usage manageable across its large tool surface.
The Verdict
Who Should Use Ghidra MCP?
Best For
- Reverse engineers and malware analysts who want an AI assistant working inside Ghidra
- Security teams automating binary triage or documentation at scale
- Researchers needing cross-version function matching across patched binaries
- Teams building CI pipelines around headless, automated binary analysis
Not Ideal For
- Users without an existing Ghidra install or reverse engineering workflow
- Teams uncomfortable granting an AI agent write access to analysis projects
- Casual users who just want a one-off decompile without setting up MCP tooling
What's Great
- Large tool surface with full write access (renaming, typing, structure creation, script execution, live debugging), not just read-only decompiles
- Batch operations and atomic transactions cut API calls and avoid partial failures during long analysis sessions
- Cross-binary documentation transfer via function hash matching speeds up re-analyzing patched versions
- Flexible deployment: GUI plugin, headless CI server, and Docker containers with optional authentication
Watch Out For
- Requires a specific Ghidra and Java version to stay compatible, so upgrades need coordination
- Write access and script execution mean a misconfigured or unsandboxed agent could corrupt analysis data
- Convention-enforcement tooling can reject renames outside its rules, which may surprise users expecting a stable API
Pricing
View all features & details
Key Features
- Function decompilation, call graph analysis, and cross-reference mapping
- Data flow analysis using PCode-graph technology
- Structure and enumeration discovery with automated field naming
- String extraction with regex search and quality filtering
- P-code emulation for isolated function execution
- Live debugger integration via Ghidra's TraceRmi framework
- Lazy tool loading so AI clients load only the tool groups they need
Use Cases
- Malware analysis and behavioral detection
- Reverse engineering of closed-source binaries
- Automated documentation workflows for large codebases
- Cross-version binary matching and annotation propagation
- API hash resolution via emulation
How It Compares
| Feature | Ghidra MCP | suidpit/ghidra-mcp | reverse-engineering-assistant |
|---|---|---|---|
| Tool count | 200+ endpoints | Smaller, focused set | Focused set |
| Write access | Full (rename, type, script, debug) | Limited | Limited |
| Deployment | GUI, headless, Docker | GUI plugin | GUI plugin |
| Batch operations | Yes, atomic transactions | No | No |