MCP Servers

A collection of Model Context Protocol servers, templates, tools and more.

B
Botquanganh MCP

MCP server by nmhuei

Created 6/4/2026
Updated about 18 hours ago
Repository documentation and setup instructions

Fallback Runner MCP

MCP server để ChatGPT có thể dùng máy của bạn kiểm tra kết nối tới các target/server CTF hoặc lab.

Repo có 2 chế độ:

  • Basic: nhẹ, dùng để test/local, có tool kiểm tra kết nối và chạy solver Python pwn/web cơ bản.
  • Full / Advanced: cài thêm Docker runner images, phù hợp chạy trên VPS hoặc máy chủ cloud.

Tool Modes

Basic Mode

Basic là chế độ mặc định (ENABLE_ADVANCED_TOOLS=false). Không cần Docker image.

Tool basic:

health_check
get_capabilities
check_target_allowed
probe_target_from_runner
run_basic_python_solver

Dùng basic khi:

  • bạn chỉ muốn ChatGPT connect tới MCP server,
  • cần kiểm tra host/port từ mạng máy bạn,
  • cần chạy solver Python nhẹ cho pwn/web qua máy bạn,
  • đang test connector,
  • chạy trên laptop/local machine.

Basic Python packages:

requests
beautifulsoup4
lxml
pwntools
pycryptodome
z3-solver
sympy
gmpy2
websocket-client
websockets

run_basic_python_solver requires a target object (host and port). That target must be allowed by ALLOWED_TCP_TARGETS. The solver receives TARGET_HOST and TARGET_PORT in its environment.

Full / Advanced Mode

Full mode bật bằng ./scripts/install_advanced_tools.sh.

Tool advanced thêm:

get_runner_environments
run_solver_fallback
validate_run_request
upload_artifact
rerun_run
get_run_log
list_recent_runs
delete_run
get_run_stdout
get_run_stderr
tail_run_output
create_workspace
upload_file_to_workspace
list_workspace_files
read_workspace_file
delete_workspace
run_command

Full mode sẽ build các Docker image nặng:

ctf-python-runner:latest
ctf-pwn-runner:latest
ctf-sage-runner:latest
ctf-forensics-runner:latest

Important Note

For normal testing, ChatGPT connector setup, and lightweight pwn/web solving, install basic only:

./scripts/install_basic.sh

Install full/advanced only on a VPS or cloud server if you need containerized solver execution, Sage, forensics, or heavier isolated tooling. The full install downloads and builds large Docker images and can take a long time on a laptop.


Quick Start: Basic

cd /home/light/Workspace/agy/botquanganh_mcp
chmod +x scripts/*.sh
./scripts/install_basic.sh

Edit .env:

ENABLE_ADVANCED_TOOLS=false
ALLOWED_TCP_TARGETS=target.host:port
BLOCK_PRIVATE_IPS=true

Start MCP server with Cloudflare Tunnel:

./scripts/start_tunnel_server.sh

The script prints a public endpoint like:

https://xxxx.trycloudflare.com/mcp

Use that URL in ChatGPT MCP Connector settings.


ChatGPT Connector Setup

  1. Run:

    ./scripts/start_tunnel_server.sh
    
  2. Copy the printed URL ending in /mcp.

  3. In ChatGPT, create/add an MCP connector.

  4. Paste the URL, for example:

    https://xxxx.trycloudflare.com/mcp
    
  5. Test with:

    health_check
    

In basic mode, ChatGPT should see exactly:

health_check
get_capabilities
check_target_allowed
probe_target_from_runner
run_basic_python_solver

Configure Allowed Targets

ALLOWED_TCP_TARGETS controls which host:port ChatGPT may probe through your machine.

Specific allowlist:

ALLOWED_TCP_TARGETS=socket.cryptohack.org:13418,example.com:443

Wildcard:

ALLOWED_TCP_TARGETS=*

Avoid * unless you understand the risk. This server uses your machine/network.

Recommended:

BLOCK_PRIVATE_IPS=true

This helps block localhost/private-network targets unless explicitly allowed for testing.


Full Install For VPS / Cloud Server

Use this only when you want ChatGPT to run solver scripts or shell commands inside isolated Docker containers.

cd /home/light/Workspace/agy/botquanganh_mcp
chmod +x scripts/*.sh
./scripts/install_advanced_tools.sh

After installation, restart the server:

./scripts/start_tunnel_server.sh

install_advanced_tools.sh will:

  • run basic install,
  • build all runner Docker images,
  • set ENABLE_ADVANCED_TOOLS=true in .env.

Manual Server Start

Without tunnel:

source .venv/bin/activate
PYTHONPATH=. python3 -m app.main

HTTP mode:

source .venv/bin/activate
PYTHONPATH=. fastmcp run app/main.py --transport http --port 8000 --host 127.0.0.1

Dev UI:

./scripts/dev.sh

Test

For local testing and lightweight pwn/web solving, basic install is enough:

./scripts/install_basic.sh
./scripts/test.sh

The test suite uses mocks for Docker paths, so it does not require full install just to validate code.


Project Layout

app/
  main.py              MCP entrypoint
  config.py            environment config
  security.py          allowlist and safety checks
  tools/
    health.py          basic health/capability tools
    probe.py           basic target connectivity probe
    basic_runner.py    basic Python pwn/web solver runner
    fallback.py        advanced solver runner tools
    workspace.py       advanced workspace tools
    runs.py            advanced run log tools
    shell.py           advanced container command tool
runner_images/         Dockerfiles for full install
scripts/
  install_basic.sh
  install_advanced_tools.sh
  start_tunnel_server.sh
  build_runner_images.sh
  test.sh
logs/                  runtime logs, ignored by git

Security Notes

  • Keep basic mode for laptop/local testing and lightweight pwn/web challenges.
  • Use full mode mainly on VPS/cloud servers with Docker.
  • Keep ALLOWED_TCP_TARGETS narrow when possible.
  • Keep BLOCK_PRIVATE_IPS=true unless you are intentionally testing local targets.
  • Full mode exposes more powerful tools, including container execution and file workspace operations.
Quick Setup
Installation guide for this server

Install Package (if required)

uvx botquanganh_mcp

Cursor configuration (mcp.json)

{ "mcpServers": { "nmhuei-botquanganh-mcp": { "command": "uvx", "args": [ "botquanganh_mcp" ] } } }