Discover Awesome MCP Servers

Hermes Plant MCP Server

Enables AI agents to call deterministic finance and quant APIs (e.g., DCF, Black-Scholes, bond pricing) with per-call payments over x402, no API keys required.

PlanTool

An MCP server that interviews users to capture project planning facts into SQLite and enforces mechanical gates to ensure structured progress.

Parakeet Transcription MCP Server

Transcribes audio/video files to text using NVIDIA's Parakeet TDT 0.6B V2 model, with optional timestamps; also retrieves model and system hardware info.

MCP Director

An intelligent orchestration server that routes user requests to appropriate specialized MCP servers and manages complex workflows.

pygdbmi-mcp-server

Enables dynamic debugging with GDB via the MCP protocol, allowing LLMs to execute GDB commands, manage breakpoints, control execution, and inspect program state.

mcp-legislation-uk

An MCP server for accessing the UK's official legislation database (legislation.gov.uk), enabling natural language queries via Pipeworx's AI gateway.

GitHub MCP Bridge

A Model Context Protocol server that enables AI agents to securely access and interact with GitHub Enterprise data, providing access to enterprise users, organizations, emails, and license information.

docsonar

A local document search MCP server that indexes folders and provides hybrid keyword+semantic search, enabling users to chat with their documents via MCP clients like Claude Desktop or Claude Code.

codex3d

Enables natural language creation and refinement of Blender scenes through structured MCP tools, with persistent object identity, visual validation, and reversible edits.

WhatsApp MCP Server

Triển khai máy chủ MCP (Message Control Protocol) của WhatsApp

Chrome Course MCP

A local MCP server that enables Codex to inspect and interact with Chrome tabs through the Chrome DevTools Protocol, primarily for collecting authorized Brightspace course materials into local folders.

acmt001-mcp

MCP server for ISO 20022 acmt.001 Account Opening (and companion acmt.* messages): message-type discovery, required-field lookup, JSON Schema introspection, IBAN/BIC/LEI validation, flat-record validation, and validated acmt XML generation.

Knowledge Graph Builder

Transforms text or web content into structured knowledge graphs using local AI models with MCP integration for persistent storage in Neo4j and Qdrant.

MCP Notion Server (@suncreation)

An MCP server that enables LLMs to interact with Notion workspaces via the Notion API, supporting page creation, database management, and content retrieval. It features markdown conversion to optimize token usage and enhanced error handling for more reliable workspace interactions.

stock-analyzer-ai-mcp

Analyze stocks with financial ratios, comparisons, and sector performance data.

MCP Template

A template MCP server built with FastMCP framework that demonstrates basic tool implementation with a simple addition calculator example.

Google Calendar MCP Server

Máy chủ Giao thức Bối cảnh Mô hình (MCP) tích hợp với API Lịch Google.

textview-mcp

Connects AI assistants to TextView for persistent memory, enabling saving, searching, and retrieving notes across sessions.

Odoo 19 MCP Server

Provides tools to interact with Odoo 19's External JSON-2 API, enabling CRUD operations and complex queries on Odoo databases with multi-company support.

AI Agent with MCP

Okay, here's a basic outline and code snippets to help you create your first MCP (Model Context Protocol) server in a Playground environment. Keep in mind that MCP is a relatively new and evolving protocol, so the specific libraries and implementations might change. This example focuses on a simplified, conceptual approach. **Conceptual Overview** 1. **What is MCP?** MCP is a protocol designed to facilitate communication between a client (e.g., an application needing a model) and a server (hosting the model). It aims to standardize how models are accessed and used, especially in distributed environments. It handles things like: * Model discovery * Model loading/unloading * Model execution (inference) * Data serialization/deserialization 2. **Simplified Approach:** For a Playground, we'll create a very basic server that: * Hosts a simple "model" (in this case, a function that adds two numbers). * Listens for requests on a specific port. * Receives data (two numbers) from a client. * Executes the "model" (adds the numbers). * Sends the result back to the client. **Code (Python - using `socket` for simplicity)** This example uses Python's built-in `socket` library for network communication. While not a full-fledged MCP implementation, it demonstrates the core concepts. ```python import socket import json # --- Model Definition (Simple Addition) --- def add_numbers(a, b): """Our "model" - adds two numbers.""" return a + b # --- Server Configuration --- HOST = '127.0.0.1' # Localhost PORT = 65432 # Port to listen on (choose a free port) # --- Server Logic --- def run_server(): with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s: s.bind((HOST, PORT)) s.listen() print(f"Server listening on {HOST}:{PORT}") conn, addr = s.accept() with conn: print(f"Connected by {addr}") while True: data = conn.recv(1024) # Receive data (up to 1024 bytes) if not data: break try: # Attempt to decode the data as JSON received_data = json.loads(data.decode('utf-8')) # Check if the data contains 'a' and 'b' keys if 'a' in received_data and 'b' in received_data: a = received_data['a'] b = received_data['b'] # Execute the "model" result = add_numbers(a, b) # Prepare the response response = {'result': result} response_data = json.dumps(response).encode('utf-8') # Send the response back to the client conn.sendall(response_data) print(f"Sent result: {result}") else: error_message = {'error': 'Invalid input format. Expected JSON with keys "a" and "b".'} conn.sendall(json.dumps(error_message).encode('utf-8')) print("Invalid input received.") except json.JSONDecodeError: error_message = {'error': 'Invalid JSON format.'} conn.sendall(json.dumps(error_message).encode('utf-8')) print("Invalid JSON received.") except Exception as e: error_message = {'error': str(e)} conn.sendall(json.dumps(error_message).encode('utf-8')) print(f"Error processing request: {e}") if __name__ == "__main__": run_server() ``` **Explanation:** * **`add_numbers(a, b)`:** This is our placeholder "model." In a real MCP server, this would be a much more complex model (e.g., a TensorFlow or PyTorch model). * **`HOST` and `PORT`:** Configure the server's address and port. `127.0.0.1` is localhost (your own machine). Choose a port that's not commonly used. * **`socket.socket(...)`:** Creates a socket object, which is the endpoint for network communication. * **`s.bind((HOST, PORT))`:** Binds the socket to the specified address and port. * **`s.listen()`:** Starts listening for incoming connections. * **`s.accept()`:** Accepts a connection from a client. This blocks until a client connects. * **`conn.recv(1024)`:** Receives data from the client (up to 1024 bytes at a time). * **`json.loads(data.decode('utf-8'))`:** Decodes the received data, assuming it's a JSON string. We're expecting the client to send a JSON object like `{"a": 5, "b": 3}`. * **`add_numbers(a, b)`:** Executes the "model" with the received data. * **`json.dumps(response).encode('utf-8')`:** Encodes the result back into a JSON string and then encodes it into bytes for sending over the network. * **`conn.sendall(response_data)`:** Sends the response back to the client. * **Error Handling:** Includes `try...except` blocks to handle potential errors like invalid JSON or other exceptions during processing. Sends error messages back to the client. **Client Code (Python)** ```python import socket import json HOST = '127.0.0.1' # The server's hostname or IP address PORT = 65432 # The port used by the server with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s: s.connect((HOST, PORT)) # Prepare the data to send (as a JSON string) data = {'a': 10, 'b': 5} json_data = json.dumps(data).encode('utf-8') s.sendall(json_data) print(f"Sent: {data}") received_data = s.recv(1024) try: response = json.loads(received_data.decode('utf-8')) print(f"Received: {response}") if 'result' in response: print(f"Result: {response['result']}") elif 'error' in response: print(f"Error: {response['error']}") else: print("Unexpected response format.") except json.JSONDecodeError: print("Received invalid JSON from server.") ``` **Explanation of Client Code:** * **`s.connect((HOST, PORT))`:** Connects to the server. * **`data = {'a': 10, 'b': 5}`:** Creates a dictionary containing the input data for the "model." * **`json.dumps(data).encode('utf-8')`:** Converts the dictionary to a JSON string and then encodes it into bytes. * **`s.sendall(json_data)`:** Sends the data to the server. * **`s.recv(1024)`:** Receives the response from the server. * **`json.loads(received_data.decode('utf-8'))`:** Decodes the JSON response. * **Prints the result or error message.** **How to Run in a Playground (e.g., Google Colab, Jupyter Notebook):** 1. **Run the Server Code:** Copy and paste the server code into a cell in your Playground and run it. It will start listening for connections. **Important:** The server will block at `s.accept()` until a client connects. 2. **Run the Client Code:** Copy and paste the client code into *another* cell in your Playground and run it. The client will connect to the server, send the data, and receive the result. **Important Considerations and Next Steps:** * **Threading/Asynchronous Operations:** The server code above is single-threaded. It can only handle one client at a time. For a real-world MCP server, you'll need to use threading or asynchronous programming (e.g., `asyncio` in Python) to handle multiple concurrent requests. * **Serialization:** MCP often uses more efficient serialization formats than JSON (e.g., Protocol Buffers, Apache Arrow). Consider using these for better performance. * **Model Management:** A real MCP server needs to handle model loading, unloading, versioning, and potentially model discovery. * **Security:** For production environments, you'll need to add security measures (e.g., authentication, authorization, encryption). * **Error Handling:** Implement robust error handling and logging. * **MCP Libraries:** Look for existing MCP libraries or frameworks in your language of choice. These will provide a more complete and standardized implementation of the protocol. However, as of late 2024, MCP is still relatively new, so library support might be limited. You might need to build parts of the protocol yourself. * **gRPC:** gRPC is a popular framework for building high-performance, language-agnostic RPC (Remote Procedure Call) systems. While not strictly MCP, it shares many of the same goals and can be a good alternative or a building block for an MCP-like system. **Example Output (in the Playground):** **Server Output:** ``` Server listening on 127.0.0.1:65432 Connected by ('127.0.0.1', <some_port_number>) Sent result: 15 ``` **Client Output:** ``` Sent: {'a': 10, 'b': 5} Received: {'result': 15} Result: 15 ``` This basic example provides a starting point for understanding the core concepts of an MCP server. To build a production-ready MCP server, you'll need to address the considerations mentioned above and potentially use more specialized libraries and frameworks. Remember to install any necessary libraries (like `protobuf` if you choose to use Protocol Buffers) using `pip install <library_name>` in your Playground environment.

MCP2Brave

Một máy chủ dựa trên giao thức MCP sử dụng API Brave cho chức năng tìm kiếm web.

macos-mcp

Local MCP server for macOS native apps: Mail, Calendar, Reminders, Notes, Messages, and Contacts. Enables reading and organizing your Mac life through a single stdio process using AppleScript/JXA.

Remote MCP Server with Bearer Auth

A Cloudflare Workers-based MCP server implementation that supports OAuth/bearer token authentication, enabling secure remote interaction with Model Context Protocol tools.

neurodivergent-memory

MCP server for knowledge graphs designed around neurodivergent thinking patterns, organizing memories into five districts with BM25 ranking and bidirectional connections.

laptop-care

A laptop maintenance agent for Claude Desktop that inspects your machine, explains findings, recommends actions, and remembers history, with safety gates enforced by the server.

LimaCharlie MCP

A local MCP server for the LimaCharlie security platform that provides investigation, administration, and content-review workflows via a broad read-only tool surface with explicit organization scoping and audit logging.

Paylocity MCP Server

Connects Claude Desktop to the Paylocity API to manage employee records, pay statements, and company headcount data through natural language. It includes automated data protection that redacts sensitive information like SSNs and bank account numbers before reaching the model.

Xiaohongshu (XHS) Creator Toolkit

An automation toolkit that enables AI-driven content publishing and creator data analysis for Xiaohongshu via the MCP protocol. It supports automated posting of image and video notes, cookie management, and performance metric tracking through natural language conversations.

@langapi/mcp-server

MCP server for AI-powered translation management in i18n projects, enabling automated locale detection, translation status checks, and sync via LangAPI.

gpt-image-mcp

An MCP server that enables Claude Code and Claude Desktop to generate and edit images using OpenAI's gpt-image-2 model, with results saved to disk.