Training small models on MCP tool capabilities? 🤖 Getting clean training data is harder than it should be 📊
The usual approach: 🔄 Run each MCP server ⚠️ Extract schemas at runtime 🚨 Deal with security risks & dependencies
Better approach: 🔍 Static analysis of .NET assemblies ✅ No code execution required 📄 Clean metadata extraction
Built McpExtract to do exactly this - analyzes assemblies instead of running potentially untrusted code. 🛠️ 🔗 https://github.com/asklar/McpExtract
How are you handling MCP training data extraction? 👇