MCPJam

MCPJam is an open-source testing and evaluation platform for people building MCP (Model Context Protocol) servers. The problem it addresses is that MCP clients do not behave identically: ChatGPT, Claude, Cursor and Copilot each read a server's tools, prompts and resources slightly differently, so a server that works in one client can quietly break in another. MCPJam exercises a server across 16 client configurations and 170+ models and reports where the behaviour diverges.

The Playground is the core surface — a chat interface that emulates real client UIs, with Chat, Trace and Raw views so you can watch every JSON-RPC message and window.openai call as it happens. It supports the OpenAI Apps SDK and MCP app UIs, and includes a DevTools-style widget emulator for iterating on widgets locally: switch between Desktop, Tablet and Mobile, and test locale changes, CSP permissions, light and dark mode, and safe-area insets without deploying.

A dedicated OAuth debugger walks through MCP authorization conformance step by step, covering the 2025-03-26, 2025-06-18 and 2025-11-25 spec versions plus the 2026-07-28 draft, with client pre-registration, Dynamic Client Registration and Client ID Metadata Documents. Alongside it sits a manual server-debugging view for tools, resources, resource templates, prompts and elicitation flows, an evals system that runs test cases with expected tool calls across multiple models to track accuracy over time, and Workspaces for sharing server configurations across a team.

It runs three ways: a hosted web app at app.mcpjam.com with no install (HTTPS servers only), a Mac and Windows desktop app, or locally via npx @mcpjam/inspector@latest with Node.js 20+, which adds support for local STDIO servers. A CLI and SDK expose the same inspection and eval primitives for scripting, and the CI/CD integration runs conformance checks and evals on every pull request in GitHub Actions or any other pipeline. The project is Apache-2.0 licensed.