Test your MCP servers with MCP Debugger

Nate Barbettini kept running into the same problem while building MCP clients: something would break, and he couldn't tell if it was his code or the server's. So he built MCP Debugger — a free browser tool that connects to any HTTP MCP server and runs close to a hundred checks against it, returning a letter grade and a detailed report of what's wrong. In this episode, we cover: - What MCP Debugger does: paste a server URL, get a grade (A through F) and a list of exactly what's missing or broken - Why he built it: debugging MCP client issues when you can't tell if the problem is the client or the server - How it uses the official MCP Conformance Suite as a foundation, plus real-world checks from his own experience building MCP tooling 100% coverage of the security and OAuth parts of the spec, 50–80% coverage of everything else, with a goal of full spec coverage - How the OAuth dance works during testing — it's semi-automated: automated checks run automatically, but OAuth flows pop a browser for real sign-in - Why it doesn't call tools yet (side effects), and where he wants to take it: interactive tool testing and tracing mode - The two main use cases: validating someone else's server before consuming it, and testing your own server while building it - Progressive disclosure in MCP servers — why server-side state hiding matters even more for agents than it does for web browsers - Why most MCP server builders aren't OAuth experts, and how a canonical validator helps them ship correct implementations without needing to be - The path toward CI integration: designing test flows once and running them as automated quality gates on every release ⏰ TIMESTAMPS: 00:00 – How OAuth testing works for MCP servers 00:33 – What MCP Debugger is and how it grades servers 01:15 – What an A, B, or failing result actually means 02:01 – Why the tool focuses on server quality across MCP clients 03:13 – How it uses the MCP Conformance suite under the hood 04:47 – What parts of the spec it covers today 06:25 – The two main use cases for MCP Debugger 10:08 – Why some checks are only semi automated 12:12 – Future plans for deeper tool and resource testing 17:40 – How to try MCP Debugger and send feedback 🔗 LINKS & RESOURCES MCP Debugger - MCP Debugger: https://mcpdebugger.dev - Nate's blog post: https://recaffeinate.co/post/introducing-mcp-debugger/ MCP Specification & Testing - MCP Conformance Suite: https://github.com/modelcontextprotocol/conformance - Model Context Protocol: https://modelcontextprotocol.io/ - MCP Inspector: https://github.com/modelcontextprotocol/inspector About the guest Nate Barbettini, Founding Engineer at Arcade - LinkedIn: https://linkedin.com/in/nbarbettini - X: https://x.com/nbarbettini MCP MVP is a series from Arcade spotlighting the people building the agentic ecosystem. New episodes dropping regularly. 💬 Join the conversation: https://discord.gg/GUZEMpEZ9p 🔧 Build with Arcade: https://arcade.dev 📖 Arcade Docs: https://docs.arcade.dev #MCPDebugger #MCP #Debugging #MCPServers #AIAgents #OAuth #AgenticWeb #DeveloperTools