Unsloth AI Documentation MCP Server

# Unsloth MCP Server - Usage Guide ## 🎉 Congratulations! You've successfully created a FastMCP server for Unsloth AI documentation! ## How to Use Your Server ### 1. Running the Server Since you've already run it successfully with: ```bash uv run python unsloth_mcp_server.py ``` The server is now ready to accept MCP client connections. ### 2. Available Tools Your server provides these 5 tools: #### 🔍 search_unsloth_docs(query: str) - **Purpose**: Search the Unsloth documentation for specific topics - **Example**: `{"query": "fine-tuning llama"}` #### 🚀 get_unsloth_quickstart() - **Purpose**: Get quickstart guide and installation instructions - **Example**: `{}` (no parameters needed) #### 🤖 get_unsloth_models() - **Purpose**: Get information about supported models - **Example**: `{}` (no parameters needed) #### 📚 get_unsloth_tutorials() - **Purpose**: Get tutorial and fine-tuning guide information - **Example**: `{}` (no parameters needed) #### 💻 get_unsloth_installation() - **Purpose**: Get detailed installation instructions - **Example**: `{}` (no parameters needed) ### 3. Testing Your Server You can test individual functions by creating a simple Python script: ```python # test_my_server.py from unsloth_mcp_server import search_unsloth_docs, get_unsloth_quickstart # Test search result = search_unsloth_docs("installation") print("Search result:", result[:200], "...") # Test quickstart result = get_unsloth_quickstart() print("Quickstart:", result[:200], "...") ``` ### 4. Connecting to MCP Clients #### With Claude Desktop Add this to your Claude Desktop config: ```json { "mcpServers": { "unsloth-docs": { "command": "uv", "args": [ "run", "python", "/home/derekanas/codes/unsloth-mcp/unsloth_mcp_server.py" ], "cwd": "/home/derekanas/codes/unsloth-mcp" } } } ``` #### With Other MCP Clients Your server uses the standard MCP stdio protocol, so it works with any MCP-compatible client. ### 5. Server Features ✅ **Web Scraping**: Fetches live content from https://docs.unsloth.ai ✅ **Content Processing**: Extracts and processes HTML content ✅ **Keyword Search**: Simple but effective keyword matching ✅ **Context Extraction**: Provides relevant sections with context ✅ **Error Handling**: Graceful error handling for network issues ### 6. Next Steps Now you can: - 🔗 Connect it to Claude Desktop or other MCP clients - 🎯 Ask questions about Unsloth AI through the MCP interface - 🔧 Extend the server with more sophisticated features - 📊 Add caching or more advanced search capabilities ### 7. Example Queries to Try Once connected to an MCP client, try asking: - "How do I install Unsloth?" - "What models does Unsloth support?" - "Show me Unsloth tutorials" - "How do I fine-tune with Unsloth?" ## 🎊 Well Done! You've successfully created a functional FastMCP server that can serve as a bridge between MCP clients and Unsloth AI documentation!