servers: MCP Server Reference for LLM Tool Access and Data Sources
servers: Bridging LLMs to Tools and Data via MCP
The servers repository offers a comprehensive collection of reference implementations for Model Context Protocol (MCP) servers, designed to equip Large Language Models (LLMs) with secure access to diverse tools and data sources. With nearly 90,000 GitHub stars, this JavaScript-based project provides concrete examples using both TypeScript and Python SDKs, illustrating how to extend LLM capabilities.
Core MCP Integration and Capabilities
At its heart, servers showcases how LLMs can securely interact with external systems through MCP. The repository categorizes its offerings into two main groups: Reference Servers and Third-Party Servers.
Reference Servers are built with the TypeScript MCP SDK or Python MCP SDK and demonstrate a wide range of MCP features. These include:
- AWS KB Retrieval: Accessing knowledge bases via Bedrock Agent Runtime.
- Brave Search: Performing web and local searches using Brave's Search API.
- EverArt: Generating AI images with various models.
- Filesystem: Secure file operations with configurable access controls.
- Git/GitHub/GitLab: Tools for reading, searching, manipulating Git repositories, managing GitHub repositories, and integrating with the GitLab API.
- Google Drive/Maps: File access and search for Google Drive, alongside location services and directions for Google Maps.
- Memory: A knowledge graph-based persistent memory system.
- Databases: Read-only access and schema inspection for PostgreSQL, SQLite, and Redis.
- Browser Automation: Utilizing Puppeteer for browser tasks.
- Fetch: Web content fetching and conversion for efficient LLM usage.
- Time Utilities: General time-related functions.
Third-Party Servers extend this ecosystem, comprising official integrations maintained by companies like 21st.dev, Adfin, AgentQL, and Aiven, which are production-ready. Community-developed servers further diversify the applications, spanning domains like blockchain, e-commerce, and productivity.
Setting Up and Configuring MCP Servers
Getting an MCP server from this repository up and running is straightforward. TypeScript-based servers can be executed directly using npx. For instance, to run the memory server:
npx -y @modelcontextprotocol/server-memory
Python-based servers leverage uvx or pip. After installing uv or pip as per their respective instructions, you can run a Python server like the Git server:
uvx mcp-server-git
Alternatively, using pip:
pip install mcp-server-git; python -m mcp_server_git
Once a server is running, you configure it into an MCP client. For clients like Claude Desktop, this involves specifying the command and its arguments within the mcpServers configuration. This step effectively wires the LLM client to the capabilities exposed by the running MCP server.