Prior Art
MCGhidra builds on the work of many people and projects. This page gives credit where it is due.
Ghidra
Section titled “Ghidra”NSA released Ghidra as open source in 2019 after years of internal development. MCGhidra would not exist without the decade of investment the agency put into building a full-featured analysis engine. The decompiler alone represents person-years of work on intermediate representations, type inference, and control flow recovery. The fact that it runs headless, supports scripting, and handles dozens of processor architectures out of the box made this project feasible.
Ghidra is available at ghidra-sre.org.
GhidraMCP by Laurie Wired
Section titled “GhidraMCP by Laurie Wired”The direct inspiration. Laurie Wired’s GhidraMCP demonstrated that connecting Ghidra to the Model Context Protocol was viable and useful. MCGhidra started as a fork of her project and evolved into a different architecture — a HATEOAS REST intermediary, multi-instance support, Docker provisioning, cursor-based pagination — but the core idea of letting an MCP agent drive Ghidra traces back to her work. The proof of concept she built made the case that this was worth pursuing further.
FastMCP
Section titled “FastMCP”FastMCP by Jeremiah Lowin is the Python framework that MCGhidra’s MCP server is built on. Its decorator-based tool registration and mixin composition pattern made it practical to organize 64+ tools into maintainable domain modules. The Context system for session isolation and progress reporting is central to how MCGhidra handles multi-client scenarios. FastMCP removed a large amount of boilerplate that would otherwise dominate the codebase.
HATEOAS and REST
Section titled “HATEOAS and REST”The Hypermedia as the Engine of Application State constraint comes from Roy Fielding’s 2000 dissertation, where he formalized the REST architectural style. Most APIs that call themselves RESTful ignore this constraint. MCGhidra embraces it because agents benefit from self-describing responses — when every result includes _links to related resources, the agent does not need to memorize URL patterns or maintain a hardcoded API map.
Model Context Protocol
Section titled “Model Context Protocol”Anthropic’s MCP specification provides the transport layer between MCGhidra and its clients. The protocol’s tool/resource/prompt abstraction maps naturally to reverse engineering workflows: tools for mutating operations like renaming symbols, resources for read-only enumeration like listing functions, and prompts for guided analysis workflows.
Related Projects
Section titled “Related Projects”MCGhidra is part of a broader ecosystem of people bridging reverse engineering tools with external interfaces. Notable related work includes:
- Binary Ninja MCP servers that expose BN’s API over the same protocol
- IDA Pro scripting bridges that have connected IDA to external tools for years
- Radare2 and rizin automation frameworks, which pioneered the idea of a scriptable RE command interface
- The growing community of MCP server authors connecting domain-specific tools to language model agents
Each of these projects informed the design decisions in MCGhidra, whether by example or by contrast. The RE tooling community has a long history of building bridges between analysis engines and the outside world — MCGhidra is one more entry in that tradition.