# Beelzebub — LLM Deception Runtime (MCP Honeypot)

> Beelzebub deploys LLM-powered decoys across SSH/HTTP/TCP/TELNET/MCP; verified 1,989★ and ships Docker Compose, Helm, and a structured CLI.

## Install

Merge the JSON below into your `.mcp.json`:

## Quick Use

```bash
git clone https://github.com/beelzebub-labs/beelzebub && cd beelzebub
docker compose build
docker compose up -d
# Alternative (Go): go build -o beelzebub . && ./beelzebub run
```

## Intro

Beelzebub deploys LLM-powered decoys across SSH/HTTP/TCP/TELNET/MCP; verified 1,989★ and ships Docker Compose, Helm, and a structured CLI.

**Best for:** Security teams building realistic decoys and testing prompt-injection / tool-call attack surfaces

**Works with:** Docker Compose, Go build, or Kubernetes via Helm; optional OpenAI/Ollama backends

**Setup time:** 10-25 minutes

### Key facts (verified)

- GitHub: 1989 stars · 190 forks · pushed 2026-05-13.
- License: GPL-3.0 · owner avatar + repo URL verified via GitHub API.
- README-backed entrypoint: `docker compose up -d`.

## Main

- Start with Docker Compose to validate the runtime loop, then switch to Helm once you know which deception services you want in production.

- Treat YAML configuration as versioned security content: iterate decoy behaviors and replay incidents with comparable configs.

- Use the CLI to validate config before rollouts, and test prompt-injection payloads only in a controlled lab.

### Source-backed notes

- README describes multi-protocol decoy services including SSH, HTTP, TCP, TELNET, and MCP.
- README provides launch paths: Docker Compose, local Go build (`./beelzebub run`), and Kubernetes via Helm.
- README documents a structured CLI and flags for config paths and memory limits.

### FAQ

- **Is this only a honeypot?**: No — README positions it as an active deception runtime with LLM-backed interactions.
- **Can I start without Kubernetes?**: Yes — Docker Compose is the quickest validation; Helm is for scaling and operations.
- **What’s a safe first test?**: Run a minimal config and test injection payloads in a controlled environment before exposure.

## Source & Thanks

> Source: https://github.com/beelzebub-labs/beelzebub
> License: GPL-3.0
> GitHub stars: 1989 · forks: 190

---

<!-- ZH -->

## Quick Use

```bash
git clone https://github.com/beelzebub-labs/beelzebub && cd beelzebub
docker compose build
docker compose up -d
# Alternative (Go): go build -o beelzebub . && ./beelzebub run
```

## Intro

Beelzebub 可在 SSH/HTTP/TCP/TELNET/MCP 上部署 LLM 驱动的诱骗服务；已验证 1,989★，并提供 Docker Compose、Helm 与结构化 CLI。

**Best for:** 需要真实诱骗与验证提示词注入/工具调用攻击面的安全团队

**Works with:** Docker Compose、Go 构建或 Helm 上 K8s；可选 OpenAI/Ollama 等后端

**Setup time:** 10-25 minutes

### Key facts (verified)

- GitHub：1989 stars · 190 forks；最近更新 2026-05-13。
- 许可证：GPL-3.0；作者头像与仓库链接均已通过 GitHub API 复核。
- README 中可对照的入口命令：`docker compose up -d`。

## Main

- 先用 Docker Compose 跑通闭环；确定生产要跑哪些 deception services 后，再迁移到 Helm/K8s。

- 把 YAML 配置当作版本化的安全内容：迭代 decoy 行为，并用可对比配置复盘事件。

- 用 CLI 在发布前校验配置；提示词注入 payload 建议只在受控环境中测试。

### Source-backed notes

- README 明确覆盖多协议诱骗服务：SSH、HTTP、TCP、TELNET 与 MCP。
- README 给出启动方式：Docker Compose、Go 本地构建（`./beelzebub run`）以及 Helm 上 Kubernetes。
- README 展示结构化 CLI：可指定配置路径与内存限制等参数。

### FAQ

- **它只是蜜罐吗？**：不止。README 将其定位为主动式 deception runtime，LLM 可生成更真实的交互。
- **不使用 Kubernetes 可以吗？**：可以。Docker Compose 适合先验证；规模化运维再考虑 Helm/K8s。
- **最安全的起步测试是什么？**：先用最小配置跑通，再在受控环境测试注入 payload，避免直接暴露。

## Source & Thanks

> Source: https://github.com/beelzebub-labs/beelzebub
> License: GPL-3.0
> GitHub stars: 1989 · forks: 190


---
Source: https://tokrepo.com/en/workflows/beelzebub-llm-deception-runtime-mcp-honeypot
Author: MCP Hub