MCP服务器集合

MCP Servers Collection

A collection of Model Context Protocol (MCP) servers built with FastMCP, each serving different APIs and data sources.

Available Servers

Last.fm MCP (lastfm_mcp/)

FastMCP server exposing the entire Last.fm API as MCP tools. Categories include:

• Artist information, similar artists, top tracks/albums
• Album metadata and tracks
• Track details, similar tracks, love/scrobble functionality
• User profiles, listening history, charts
• Global charts, geo-specific data
• Tag-based content and search

Requirements: Last.fm API key (free from https://www.last.fm/api)

Grokipedia MCP (grokipedia_mcp/)

FastMCP server for scraping and structuring content from Grokipedia pages.

Features:

• Scrape page content into structured sections with headings and text blocks
• Web scraping using BeautifulSoup
• Clean separation between scraping logic and MCP server

Yahoo Finance MCP (yfinance_mcp/)

FastMCP server exposing Yahoo Finance data as MCP tools. No API key required.

• Real-time quotes, market summary, market status
• Historical OHLCV data with configurable intervals
• Symbol search
• Company profiles, income statements, balance sheets, earnings
• Analyst price targets and recommendations
• Options chains and news

Web MCP (web_mcp/)

FastMCP server for web search and URL fetching. No API key required.

• Web search via a local self-managed SearXNG instance (auto-installs on first run)
• News search with time filtering
• URL fetch with content extraction (Markdown, text, or raw HTML)
• Queries multiple engines simultaneously: Google, Bing, Brave, DuckDuckGo, Startpage, and more

Android MCP (android_mcp/)

FastMCP server for controlling an Android device via ADB. Screenshots are served over HTTPS for vision model analysis.

• Shell commands, UI interaction (tap, swipe, type, key press), UI hierarchy dump
• App management (install, uninstall, launch, stop, list, clear data)
• Screenshot capture with compression, served via built-in HTTPS file server
• Device info, file push/pull, screen wake/state

Requirements: adb and openssl on PATH, connected Android device with USB debugging enabled

Shell MCP (shell_mcp/)

FastMCP server for shell execution and file operations over SSH. Optionally auto-provisions a VirtualBox VM (Debian 13) as an isolated sandbox.

• Execute commands, read/write files, upload/download via SFTP
• List directories, get system info (memory, disk, uptime)
• Fetch generated files via a local HTTP URL
• Auto-creates and manages a Debian VM if VirtualBox is installed; otherwise connects to any SSH host

Requirements: SSH-accessible target, or VirtualBox with vboxmanage on PATH

Docker Shell MCP (docker_shell_mcp/)

Docker-backed version of Shell MCP with the same command and file tools, using an auto-managed Ubuntu container instead of SSH or VirtualBox.

• Fresh container and workspace volume every server start by default (DOCKER_EPHEMERAL=false to persist instead), matching the clean-slate-per-session behavior of shell_mcp/webshell_mcp
• Configurable CPU, memory, PID, user, and network settings
• Auto-builds and caches the sandbox image on first launch
• Copies files through the Docker CLI and serves generated files over local HTTP

Requirements: Docker CLI and a running Docker daemon

Subagent MCP (subagent_mcp/)

FastMCP server for delegating tasks to local LLM sub-agents via LM Studio or any OpenAI-compatible endpoint.

• Named agents with persistent system prompts, per-agent model selection, and MCP server access
• Agentic tool-use loop: agents can use tools from other MCP servers (web search, shell, Android, etc.)
• Parallel execution across multiple models or agents with auto-calibrated concurrency
• Multi-turn conversations with history tracking
• MCP server pool with .env pre-registration and runtime management

Requirements: OpenAI-compatible endpoint with Responses API support (e.g. LM Studio)

Colab Shell MCP (colab_shell_mcp/)

Shell execution and file operations inside a Google Colab runtime, with the same tool surface as shell_mcp/docker_shell_mcp/webshell_mcp but backed by Colab's free CPU/GPU/TPU VM.

• Runs directly inside Colab (local backend) or drives a remote Colab from your laptop through a bundled stdlib HTTP bridge over a tunnel (remote backend)
• Execute commands, read/write files, upload/download, system info (including the attached GPU), and serve generated files over local HTTP
• COLAB_MODE=auto picks the backend from whether a bridge URL is configured

Requirements: A Google Colab runtime (local backend needs nothing extra; remote backend needs the bridge running in Colab plus a tunnel)

WebShell MCP (webshell_mcp/)

Combined FastMCP server: sandboxed shell execution + web search and fetching, all in one VM. SearXNG runs inside the VM as a systemd service.

• All shell tools: execute commands, read/write files, upload/download via SFTP, system info, file serving
• All web tools: SearXNG web/news search, URL fetch with content extraction
• Auto-provisions a dedicated VM (ai-webshell) with SearXNG installed and running on first launch

Requirements: VirtualBox with vboxmanage on PATH, or an SSH target with SearXNG already running

Fire TV MCP (firetv_mcp/)

FastMCP server for controlling Amazon Fire TV devices via ADB over WiFi.

• D-pad navigation, media playback control, volume, now-playing info
• App management with 30+ friendly aliases (netflix, prime, disney+, etc.), deep linking, sideloading
• Content search, open settings sections directly
• Screenshot capture served via HTTPS, screen recording, UI hierarchy dump
• Device info, network info, storage, brightness, bluetooth, reboot
• Auto-discovers Fire TV on the network if no IP is configured

Requirements: adb and openssl on PATH, ADB debugging enabled on Fire TV

RapidAPI MCP (rapidapi_mcp/)

Suite of FastMCP servers wrapping the RapidAPI integrations used in the Tyumi project. Each domain (jobs, finance, food, entertainment, social, real estate, news, search) runs as a separate FastMCP instance.

Requirements: RapidAPI key (RAPIDAPI_KEY environment variable or .env)

Common Setup

Each server follows a similar structure:

package_name/
├── pyproject.toml      # Modern Python packaging
├── requirements.txt    # Dependencies
├── server.py          # FastMCP server and tool definitions
├── README.md          # Server-specific documentation
└── client_lib/        # Core business logic
    ├── __init__.py
    └── client.py      # API/scraping implementation

Running Servers

# Example for Grokipedia MCP
cd grokipedia_mcp
python -m venv .venv && source .venv/bin/activate  # On Windows: .venv\Scripts\activate
pip install -r requirements.txt
python server.py  # Launches MCP server for client integration

See individual server READMEs for specific setup instructions and configuration.

Development

• Built with FastMCP for MCP compatibility
• Clean architecture with separation of concerns
• Python 3.9+ required
• Each server can be used as a standalone package
• Pull requests and pushes to main test each packaged MCP server independently in GitHub Actions