Categories
Developer Tools AI Agents CI/CD Model Context Protocol LLM Evaluation API Testing MCP Testing OAuth DebuggingStars
Forks
Watchers
Developer links
MCPJam Inspector
Every AI client reads a Model Context Protocol (MCP) server a little differently, and MCPJam Inspector is the testing platform that shows exactly where yours breaks before your users find out. Connect a remote HTTP server or a local STDIO one, then send a single prompt to Claude, ChatGPT, and Cursor panes side by side and compare answers, tool calls, token counts, and latency in Chat, Trace, and Raw views. MCP App widgets render inline in a device emulator with locale, light and dark mode, and safe-area controls. The OAuth Debugger walks authorization one request at a time across the 2025-03-26, 2025-06-18, 2025-11-25, and draft 2026-07-28 spec versions, with dynamic client registration, pre-registered clients, and client ID metadata documents. Conformance runs score protocol, apps, tasks, and OAuth checks in one report, and Evals turn expected tool calls into repeatable test cases that models must pass. Run tools, resources, prompts, and elicitation flows by hand with every JSON-RPC message logged, script the same checks from the CLI and SDK, and gate pull requests in GitHub Actions or GitLab CI. Chat uses free hosted models or your own OpenAI, Anthropic, Google, xAI, Groq, or Mistral keys. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. Apache-2.0 licensed with an enterprise-licensed evals and services module.
Benefits
- See how each AI client behaves
- Claude, ChatGPT, and Cursor panes answer the same prompt side by side, so differences in tool choice, arguments, latency, and token use show up before a user reports a bug.
- Find the exact failing OAuth step
- The guided debugger runs MCP authorization request by request, from the unauthenticated 401 through resource metadata and authorization server discovery, and explains each exchange with the raw HTTP behind it.
- Catch regressions before they ship
- Eval suites assert which tools fire with which arguments, optionally add an LLM judge, and run across models, while the CLI can fail GitHub Actions or GitLab CI pull requests on the result.
- Probe servers without any model key
- Tools, Resources, Prompts, the OAuth Debugger, and conformance checks talk only to your MCP server, so protocol testing needs no LLM provider key, and Playground chat can use free hosted models.
Features
- Multi-client Playground
- Chat, Trace, and Raw views over Claude, ChatGPT, and Cursor client emulations, with inline MCP App and OpenAI Apps SDK widgets and a Chrome DevTools-style widget emulator.
- Guided OAuth debugger
- Step-by-step authorization diagrams with Guide and Raw views for spec versions 2025-03-26 through 2025-11-25 and the 2026-07-28 draft, covering DCR, client pre-registration, and CIMD.
- Conformance reports
- Protocol, Apps, Tasks, and OAuth suites score a server against the specification and list every check with its timing, outcome, and a link to the source requirement.
- Evals and CI/CD
- Test cases with expected tool calls and optional LLM-as-judge run across models, track accuracy over time, and execute in GitHub Actions, GitLab CI, or any pipeline.
- CLI and SDK
- The @mcpjam/cli and @mcpjam/sdk packages probe servers, run doctor and OAuth checks, snapshot capabilities, and assert on tool and resource shapes from your own test suites.
- Manual server inspector
- Run tools, resources, templates, prompts, and elicitation flows by hand with full JSON-RPC logs, connecting HTTP servers or local STDIO commands with custom headers.