发现优秀的 MCP 服务器

CereBro

一个与模型无关的、用于 .Net 的 MCP 客户端-服务器

Adobe Learning Manager MCP Server

Enables LLM clients to interact with Adobe Learning Manager, including user management, learning object browsing, enrollment management, badge checking, and async job execution via the ALM Prime API.

resend-email-mcp

The most complete Resend MCP server — full coverage of the Resend API (emails, domains, contacts, broadcasts, templates, segments, topics, webhooks, logs) plus a unique debug/diagnostics layer no other Resend MCP offers: deliverability analysis, DNS troubleshooting, email lifecycle inspection, bounce explanation, and account auditing. Works with Claude Code, Cursor, Claude Desktop, and any other

INSEE MCP Server

Enables searching companies in the SIRENE database using SIREN, SIRET, or company name via the INSEE API.

🚀 MCP File System API

Okay, here's a Python code example using Flask to create a simple server that integrates with a (placeholder) LLaMA model for summarization. This example focuses on the structure and integration points. **Important considerations and placeholders are marked with comments.** ```python from flask import Flask, request, jsonify import torch # Import PyTorch #import llama # Placeholder: Replace with your actual LLaMA model import #from transformers import AutoTokenizer, AutoModelForSeq2SeqLM # Alternative for Hugging Face models app = Flask(__name__) # --- Model Loading and Setup --- # This section is CRUCIAL and needs to be adapted to your specific LLaMA model. # Option 1: If you have a custom LLaMA implementation # model = llama.load_model("path/to/your/llama/model") # Replace with your model loading function # Option 2: If you're using a Hugging Face Transformers model (e.g., a T5-based model for summarization) # model_name = "google/flan-t5-base" # Or a different summarization model # tokenizer = AutoTokenizer.from_pretrained(model_name) # model = AutoModelForSeq2SeqLM.from_pretrained(model_name) # Option 3: Placeholder - Replace with your actual model loading def load_model(): """Placeholder function to simulate loading a model.""" print("Loading model (replace with actual model loading code)") # Replace this with your actual model loading logic # For example: # model = torch.load("path/to/your/model.pth") # model.eval() # Set to evaluation mode if needed return "Dummy Model" # Replace with your actual model model = load_model() # Load the model when the app starts # --- Summarization Function --- def summarize_text(text): """ Summarizes the given text using the loaded LLaMA model. Args: text: The input text to summarize. Returns: The summarized text. """ print(f"Summarizing text: {text[:50]}...") # Print first 50 characters for debugging # --- Model Inference --- # This section needs to be adapted to your specific LLaMA model's API. # Option 1: Custom LLaMA model # summary = model.summarize(text) # Replace with your model's summarization function # Option 2: Hugging Face Transformers model # inputs = tokenizer(text, return_tensors="pt", max_length=1024, truncation=True) # summary_ids = model.generate(inputs.input_ids, max_length=150, min_length=40, length_penalty=2.0, num_beams=4, early_stopping=True) # summary = tokenizer.decode(summary_ids[0], skip_special_tokens=True) # Option 3: Placeholder - Replace with your actual model inference code summary = f"Placeholder Summary for: {text[:20]}..." # Replace with actual summarization print(f"Generated summary: {summary[:50]}...") # Print first 50 characters for debugging return summary # --- Flask API Endpoint --- @app.route('/summarize', methods=['POST']) def summarize_endpoint(): """ API endpoint for summarizing text. Expects a JSON payload with a 'text' field. """ try: data = request.get_json() text = data.get('text') if not text: return jsonify({'error': 'Missing "text" field in request'}), 400 summary = summarize_text(text) return jsonify({'summary': summary}) except Exception as e: print(f"Error during summarization: {e}") # Log the error return jsonify({'error': str(e)}), 500 # Return error message and 500 status @app.route('/health', methods=['GET']) def health_check(): """Simple health check endpoint.""" return jsonify({'status': 'ok'}) if __name__ == '__main__': app.run(debug=True, host='0.0.0.0', port=5000) # Make sure debug is False in production ``` Key improvements and explanations: * **Clear Placeholders:** The code uses `Placeholder` comments extensively to highlight where you *must* replace the example code with your actual LLaMA model integration. This is the most important part. * **Error Handling:** Includes `try...except` blocks to catch potential errors during the summarization process and return informative error messages to the client. This is crucial for debugging and production stability. * **Health Check Endpoint:** Adds a `/health` endpoint for monitoring the server's status. This is essential for deployment and monitoring. * **Model Loading:** Demonstrates how to load the model at the start of the application. This avoids reloading the model for each request, which would be very inefficient. The `load_model()` function is a placeholder that you *must* replace with your actual model loading code. * **Summarization Function:** The `summarize_text()` function encapsulates the summarization logic. This makes the code more modular and easier to test. Again, the model inference part is a placeholder. * **Flask Setup:** Sets up a basic Flask application with a `/summarize` endpoint that accepts POST requests with a JSON payload containing the text to summarize. * **JSON Handling:** Uses `request.get_json()` to properly parse the JSON payload from the request. * **Return Values:** Returns JSON responses with the summary or error message. * **Logging:** Includes `print` statements for debugging. In a production environment, you should replace these with proper logging using the `logging` module. * **Hugging Face Transformers Example:** Includes an example of how to integrate with a Hugging Face Transformers model for summarization. This is a common way to use pre-trained models. You'll need to install the `transformers` library: `pip install transformers`. * **`host='0.0.0.0'`:** This makes the server accessible from outside the local machine. Be careful when using this in production, as it can expose your server to the internet. * **`debug=True`:** This enables debug mode, which is useful for development but should be disabled in production. * **Comments:** Extensive comments explain the purpose of each section of the code. **How to Use:** 1. **Install Flask:** `pip install Flask` 2. **Install PyTorch:** `pip install torch` (if you are using PyTorch) 3. **Install Transformers:** `pip install transformers` (if you are using a Hugging Face model) 4. **Replace Placeholders:** **This is the most important step.** Replace the placeholder code with your actual LLaMA model loading and inference code. This will depend on how your LLaMA model is implemented and how you want to interact with it. 5. **Run the Application:** `python your_script_name.py` 6. **Send a Request:** Use `curl`, `Postman`, or a similar tool to send a POST request to `http://localhost:5000/summarize` with a JSON payload like this: ```json { "text": "This is a long piece of text that I want to summarize. It contains many sentences and paragraphs. The goal is to reduce the text to its most important points." } ``` **Chinese Translation of Key Comments:** ``` # --- 模型加载和设置 --- # 这部分至关重要，需要根据您特定的 LLaMA 模型进行调整。 # --- 模型推理 --- # 这部分需要根据您特定的 LLaMA 模型的 API 进行调整。 # 替换占位符：这是最重要的步骤。将占位符代码替换为您实际的 LLaMA 模型加载和推理代码。这将取决于您的 LLaMA 模型是如何实现的，以及您希望如何与它交互。 ``` **Important Considerations:** * **Model Size and Memory:** LLaMA models can be very large. Make sure you have enough memory to load and run the model. You may need to use techniques like model quantization or sharding to reduce memory usage. * **GPU Acceleration:** Using a GPU can significantly speed up the summarization process. Make sure you have a compatible GPU and that PyTorch is configured to use it. * **Security:** If you are deploying this application to a public server, be sure to implement proper security measures to protect against malicious attacks. * **Rate Limiting:** Implement rate limiting to prevent abuse of the API. * **Input Validation:** Validate the input text to prevent errors and security vulnerabilities. * **Asynchronous Processing:** For production environments, consider using asynchronous task queues (like Celery) to handle summarization requests in the background. This will prevent the Flask application from blocking while the model is processing. * **Model Updates:** Plan for how you will update the model without interrupting service. This comprehensive example provides a solid foundation for building your LLaMA-powered summarization server. Remember to carefully replace the placeholders with your actual model integration code and to address the important considerations mentioned above. Good luck!

localmem-mcp

Gives AI agents persistent, local-first memory using SQLite and on-device embeddings, enabling semantic search and recall across sessions with no cloud calls.

Asana Universal MCP Server

Provides a standardized interface for interacting with Asana's tools and services through a unified API, enabling AI assistants to manage tasks, projects, and workflows.

Disco

Superhuman data-driven science. Allows agents to upload any tabular dataset, specify a target column, and get validated predictive patterns (with p-values, effect sizes, and context from literature) that surface feature interactions and subgroup effects you'd otherwise miss. Many discoveries already made, free for open data!

Obsidian MCP Server

Enables AI assistants to manage and search Obsidian notes, folders, metadata, and links directly.

Obsidian Tools MCP Server

Enables comprehensive management of Obsidian vaults with full CRUD operations, advanced search, link/tag extraction, backlinks discovery, frontmatter editing, and template-based note creation through natural language.

Creative Ideation MCP Server

Utilizes the Google Gemini API to generate context-specific categories and diverse options for creative brainstorming. It features a random sampling function to help users bypass predictable AI patterns and discover unexpected, innovative ideas.

nseap-mcp

Enables local deliverable checks and a submission-review-notification workflow for the NSEAP education platform via stdio MCP, integrating with Cogseed.

DAZ Command MCP Server

An MCP server that provides session-based shell command execution and file management with intelligent LLM-powered summarization. It enables users to manage isolated workflows and track progress through automated event logging and session history.

Meticulous Espresso MCP Server

Connects your Meticulous espresso machine to an LLM, enabling recipe generation from natural language, shot analysis, grinder dial-in, and a persistent shot diary through Claude.

mcp-server-email MCP server

I am a language model, and I don't have an email address. I am accessed through interfaces provided by Google.

MCP Skills Manager

A server that serves reusable markdown skills (instructions, playbooks, references) to agents, with a web UI for editing and organizing skills, profiles, and both HTTP and stdio transports.

prometheus-mcp-server

A TypeScript-based MCP server that enables users to interact with Prometheus metrics using PromQL queries and discovery tools. It allows LLMs to retrieve time-series data, metadata, alerts, and system status directly from a Prometheus instance.

Minecraft RCON MCP Server

Connects AI agents to Minecraft servers via RCON to execute commands, monitor logs, and perform read-only SQLite database queries. It is specifically designed to facilitate AI-assisted plugin development, live debugging, and automated testing workflows.

SailPoint MCP Server (Authless)

Provides a template for deploying a remote MCP server on Cloudflare Workers without authentication, allowing customization of tools and connection to AI Playground or Claude Desktop.

mcp-openai-platform

A local MCP server exposing the OpenAI platform REST API as tools for file management, fine-tuning, inference, images, audio, batch processing, and organization usage/costs.

Pump.fun MCP Server

A Cloudflare Worker that implements the MCP Streamable HTTP transport for AI agents to interact with pump.fun on Solana.

Clear Thought MCP Server

Provides systematic thinking tools including mental models, design patterns, debugging approaches, and collaborative reasoning frameworks to enhance problem-solving and decision-making capabilities.

blumira-mcp

An MCP server for Blumira SIEM platform, enabling management of security event detection, alerts, and threat response through Blumira's API.

Livepeer Agent MCP Server

A terminal-first Livepeer client that provides identity, access, and email verification through the MCP protocol, enabling interaction with Daydream AI models and browser automation.

mcp-drizly

An MCP server that enables AI assistants to search for alcohol products, manage shopping carts, and place orders on Drizly. It uses browser automation to check product availability, discover local stores, and track delivery status.

BlenderMCP

Connects Claude AI to Blender through the Model Context Protocol, enabling AI-assisted 3D modeling, scene creation, object manipulation, and material control. Supports downloading assets from Poly Haven and generating 3D models through Hyper3D Rodin.

pg-health

Archived MCP server for PostgreSQL health monitoring; functionality merged into pg-dash.

Quilt

Quilt gives every AI agent line-level authorship in a shared Git checkout, so multiple agents work in one repo, and each commits only its own lines, with symbol-level claims that block collisions before any bytes change.

Vps-mcp

Enables a ChatGPT Web conversation to act as a coding agent on a VPS, with project-bound chats, Git worktree isolation, live activity streaming, structured questions, and file/image exchange via MCP.