REA MCP Setup: Connect Reverse Engineering to Your AI Agent

MCP in one paragraph

Model Context Protocol (MCP) is an open standard that lets AI assistants call external tools and data sources. For reverse engineering, MCP is the bridge between the agent (Claude Code, Codex, Cursor, Gemini CLI…) and the engine that actually reads binaries (Ghidra, Hopper, IDA). Without a bridge, your agent can only talk about code it can see. With one, it can open a binary, trace a function, and show you evidence.

REA is that bridge — one MCP server for many targets and engines.

How REA registers with your agent

The setup command does the wiring for you:

npx rea-agents setup

It registers REA's MCP server with each agent you select and installs matching workflow instructions. Your existing configuration is backed up first, and you review every proposed change before approving it. After setup, restart your agent so it picks up the new server.

Manual MCP registration is also possible for any MCP-capable client — see the installation and setup guide in the REA repository.

Engine providers: Ghidra, Hopper, IDA

REA supports several providers for native binary analysis, and you choose per your installation:

• Ghidra — works with an existing local Ghidra installation. For large binaries, raise the provider's startup deadline with REA_GHIDRA_STARTUP_TIMEOUT_MS. Headless analysis and 16-bit DOS analysis are supported.
• Hopper — an existing Hopper installation is used if present; setup can optionally install Hopper with your approval.
• IDA — works with an existing IDA installation, including an optional native MCP bridge. See the IDA guide in the repository.
• No engine needed for static JavaScript and Electron analysis, which REA handles on its own.

On Linux, native analysis and firmware tool launches need a procps-compatible ps on REA's PATH — that's a process-ownership check, and the fix is simply installing procps if it's missing.

Other MCP servers in the reversing space

REA is not alone — "MCP for reverse engineering" is becoming an ecosystem:

• WinDbg MCP — Microsoft ships an official MCP server with recent WinDbg versions for Windows debugging and crash-dump analysis
• JEB MCP — the commercial Android decompiler JEB exposes official MCP capabilities
• hexvero — a static triage tool for Electron archives and JavaScript bundles with a built-in MCP server
• IDA MCP bridges — community and native bridges let agents drive IDA sessions

The pattern is the same everywhere: the agent reasons, the engine executes, MCP connects them.

Troubleshooting the common cases

• The agent can't see REA's tools — restart the agent after setup. MCP servers are loaded at session start.
• Ghidra analysis times out on a large binary — increase REA_GHIDRA_STARTUP_TIMEOUT_MS.
• Native analysis fails on Linux — install procps so a compatible ps is available.
• You want to verify what setup changed — the setup step backs up your previous agent configuration; review the backups it reports.

Working on a specific target type? The main REA guide covers installation end to end, and the tools directory lists the engines worth wiring up.