@mcpjams launched WebMCP Inspector 🌐
we’ve integrated a browser into our chat client so you can test agent workflows that require browsing + we’ve added support for WebMCP debugging
in ~20 min, I’ll walk through:
• patterns from testing leading MCP servers: where agents fail going from user intent to user value
• where context belongs in a stateless MCP server
• how to build a testing ground where you can see the user’s intent, the agent’s actions, and the outcome across AI clients
• how to turn user journeys into tests across clients, so you can continuously catch breaking user outcomes before you ship
speaking TODAY at MCP Community Connect at 📍 GitHub HQ in San Francisco @ ⏲️ 3:30PM!
How do you know your MCP 2.0 server is delivering value to your users?
🎤 "Effective MCP: Building Reliable Stateless MCP Servers"
i have my differences with this platform, but i appreciate that efforts like these can just happen on weekends.
"despite all the progress, we still only understand a tiny fraction of what goes on inside these models"
surreal that one of our best inventions is still a blackbox to the people building it. coordinating on the research side is essential. we need to attempt to understand this thing together.
@mcpjams launched WebMCP Inspector 🌐
we’ve integrated a browser into our chat client so you can test agent workflows that require browsing + we’ve added support for WebMCP debugging
speaking at @AgenticAIFdn's AGNTCON+MCPCon in October!
my talk: Rethinking CI/CD Release Gates for Agent-native Software
how do you know your software is "good" and ready-to-ship for agents consuming it?
error handling in MCP is so important pt. 1
Used Atlassian's MCP server to query a project (that didn't exist), it returned an empty response: { "issues": [], "isLast": true }
Claude said the project had no issues and offered to create a bunch in it