# Ghidra MCP

> Use when working against a local GhidraMCP GUI or headless server, especially for binary import, metadata verification, analysis, export checks, or raw endpoint access.

- Skill: `maleick/ghidra-mcp` (Agent Skill, multi-file: 4 files)
- Install (CLI): `npx skillmds@latest add maleick/ghidra-mcp`
- Raw SKILL.md: https://api.skillmd.com/api/skills/maleick/ghidra-mcp/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: Maleick (https://skillmd.com/u/maleick)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/maleick/ghidra-mcp

---


# Ghidra MCP

## When to use
- Verify that a local GhidraMCP server is reachable.
- Load binaries into the headless Docker server and confirm metadata or analysis output.
- Call raw HTTP endpoints without retyping `curl` commands.
- Work around headless-specific constraints such as container-only file visibility.

## Current local workflow
The common local setup in this environment is:
- headless server on `http://127.0.0.1:8089`
- Docker container named `ghidra-mcp`
- writable container data path `/data`

Canonical project location:

```bash
/Users/maleick/Projects/ghidra-mcp-skill
```

The local skills path points here via symlink:

```bash
/Users/maleick/.codex/skills/ghidra-mcp
```

Use the bundled helper:

```bash
python3 /Users/maleick/Projects/ghidra-mcp-skill/scripts/ghidra_mcp.py check
```

Documented surface for this skill version:
- total catalog: `193`
- GUI endpoints: `175`
- headless endpoints: `183`
- validated against the `ghidra-mcp` `v5.1.0` catalog in `references/endpoints.json`

Verify the live runtime against the catalog instead of assuming parity from the upstream release notes:

```bash
python3 /Users/maleick/Projects/ghidra-mcp-skill/scripts/ghidra_mcp.py parity --mode headless
```

## Recommended headless sequence
1. Verify connectivity.
2. Copy the host binary into the container if the repo path is not mounted.
3. Load the container-local path with `/load_program`.
4. Read `/get_metadata`.
5. Run `/run_analysis` if needed.
6. Query exports, imports, functions, or any raw endpoint.

There is no special bulk-upload endpoint. For more than one executable, repeat the same container copy and `/load_program` flow per file, then qualify later calls with `program=<basename>` when more than one program may be loaded.

One-command verification:

```bash
python3 /Users/maleick/Projects/ghidra-mcp-skill/scripts/ghidra_mcp.py verify \
  /absolute/path/to/binary.exe --analyze
```

Multi-binary headless verification:

```bash
python3 /Users/maleick/Projects/ghidra-mcp-skill/scripts/ghidra_mcp.py verify-many \
  /absolute/path/to/eqgame.exe \
  /absolute/path/to/eqmain.dll \
  --analyze
```

## Raw endpoint access
For endpoints not wrapped by the helper yet, use:

```bash
python3 /Users/maleick/Projects/ghidra-mcp-skill/scripts/ghidra_mcp.py call \
  POST /load_program --form file=/data/eqgame.exe

python3 /Users/maleick/Projects/ghidra-mcp-skill/scripts/ghidra_mcp.py call \
  GET /get_metadata --query program=eqgame.exe
```

POST example:

```bash
python3 /Users/maleick/Projects/ghidra-mcp-skill/scripts/ghidra_mcp.py call \
  POST /run_analysis
```

Program-qualified examples for the common headless sequence:

```bash
python3 /Users/maleick/Projects/ghidra-mcp-skill/scripts/ghidra_mcp.py call \
  GET /list_functions --query program=eqgame.exe --query limit=20

python3 /Users/maleick/Projects/ghidra-mcp-skill/scripts/ghidra_mcp.py call \
  GET /decompile_function --query program=eqgame.exe --query address=0x401000

python3 /Users/maleick/Projects/ghidra-mcp-skill/scripts/ghidra_mcp.py call \
  GET /get_metadata --query program=eqgame.exe
```

## Headless caveats observed here
- Host paths are not visible inside the Docker container unless the repo is mounted there. If needed, copy binaries into `/data` first.
- Treat `/mcp/schema` as the source of truth for what the current runtime exposes.
- `server/status` refers to optional Ghidra Server integration, not whether the HTTP API itself is up.
- Some endpoints are documented for only one runtime mode even if the other mode exposes a placeholder or error stub.

## Output expectations
- Prefer reporting:
  - endpoint reachability
  - live mode and schema count from `/get_version` and `/mcp/schema`
  - binary path used inside the container
  - metadata summary: image base, memory size, function count
  - whether analysis ran successfully
  - any headless limitation that blocks the next step

## Source
The upstream repo advertises a large API surface and both GUI/headless modes:
- [bethington/ghidra-mcp](https://github.com/bethington/ghidra-mcp)