MalBuddy
MCP server that integrates a headless Ghidra instance for binary reverse-engineering, enabling automated analysis and decompilation through a standardized interface.
README
Dummy Agent Malware
Reverse-engineering assistant built with Google ADK and a vendored Ghidra MCP stack
(Docker headless on :8089 → SSE bridge on :8081 → ADK agent).
Prerequisites
- Docker (for the headless Ghidra MCP server)
- Python 3.10+
- LLM endpoint — an OpenAI-compatible
/v1/chat/completionsserver (e.g. LiteLLM proxy) athttp://localhost:4000/v1
Setup
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
Ensure ghidra-mcp/docker/.env has a non-empty GHIDRA_MCP_AUTH_TOKEN
(a shell export alone is not enough for Docker Compose).
./scripts/start_ghidra_docker.sh creates/fills .env from .env.example,
reuses a shell GHIDRA_MCP_AUTH_TOKEN if set, or generates one with
openssl rand -hex 32.
Start Ghidra Docker
Headless Ghidra MCP HTTP API on :8089:
./scripts/start_ghidra_docker.sh
Start MCP bridge
SSE bridge on :8081 (leave this running in its own terminal):
./scripts/start_ghidra_bridge.sh
The bridge loads GHIDRA_MCP_AUTH_TOKEN from ghidra-mcp/docker/.env and
talks to http://127.0.0.1:8089.
Verify stack
With Docker and the bridge running:
pytest tests/test_ghidra_stack.py -v
These are integration smoke tests (@pytest.mark.integration). They skip cleanly
if services are down; they fail if only half the stack is up.
Start agent
From the parent directory of this folder:
adk web
Open the ADK web UI and select Dummy_Agent_Malware.
agent.py expects the bridge at http://127.0.0.1:8081/sse (GHIDRA_MCP_URL).
Config knobs
| Setting | Where |
|---|---|
| LLM base URL / API key / model | agent.py (LLM_BASE_URL, LLM_API_KEY, LLM_MODEL_NAME) |
| Bridge SSE URL for ADK | agent.py (GHIDRA_MCP_URL) |
| Ghidra HTTP auth token | ghidra-mcp/docker/.env (GHIDRA_MCP_AUTH_TOKEN) |
Troubleshooting
- Docker not healthy / connection refused on :8089 — wait for the container
healthcheck (
docker compose -f ghidra-mcp/docker/docker-compose.yml ps), check logs withdocker logs ghidra-mcp, and confirm the token in.envmatches what compose started with. - Missing / empty auth token — put
GHIDRA_MCP_AUTH_TOKEN=...inghidra-mcp/docker/.env(not only in your shell). Re-run./scripts/start_ghidra_docker.sh(it can copy a shell token into.envor generate one), then recreate the container. - Bridge connection refused on :8081 — start
./scripts/start_ghidra_bridge.shand ensure Docker is already up; the bridge needs:8089plus the same token.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。