MCPJam Inspector

Visit Website

Introduction

MCPJam describes itself as the open-source testing and evaluations platform for MCP server developers who ship. You can interactively test tools, prompts, resources and authorisation across 16 client configurations and more than 170 models, run evals on test cases, see what is breaking and gate regressions before deployment. It runs hosted or locally for HTTP/S and local stdio servers.

Its key insight is that ChatGPT, Claude, Cursor and other clients each read your server differently. MCPJam catches those differences, shows where your server breaks and tells you how to fix it. Cross-client evals score agent behaviour across client configurations, including ChatGPT, Claude, Cursor and Copilot, tracking accuracy, latency and tool-call performance over time.

For hands-on debugging, the inspector shows your server’s tools, resources and prompts, lets you test prompts with real LLMs and traces every JSON-RPC message and OAuth exchange, with Chat, Trace and Raw views. It supports the OpenAI Apps SDK and MCP app UIs, with a Chrome DevTools-style widget emulator: invoke a tool to render its widget, watch window.openai messages and test locale changes, CSP permissions, light and dark mode, hover and touch, and safe-area insets.

Authorisation is a frequent source of MCP bugs, and MCPJam’s OAuth debugger runs guided conformance checks with step-by-step explanations against every version of the MCP OAuth spec (2025-03-26, 2025-06-18, 2025-11-25 and the 2026-07-28 draft), supporting client pre-registration, Dynamic Client Registration and Client ID Metadata Documents.

Everything the original inspector does is covered too: tools, resources, resource templates, prompts and elicitation flows with full JSON-RPC observability. A playground supports Skills; local skills are read from your filesystem and never leave your machine. A CLI, an SDK for testing MCP servers in code and CI/CD actions let you gate every PR on model behaviour.

MCP servers are software like any other, but they have an unusual consumer: a model that reads tool names and descriptions and decides what to call. That makes two kinds of testing necessary: protocol-level checks that requests, responses, auth and errors behave correctly, and behavioural checks that real models actually choose and use the tools as intended. MCPJam is one of the few tools that tackles both halves in one place.

The source code is at MCPJam/inspector on GitHub under the Apache-2.0 (with its eval services under a separate licence) licence, with about 2,300 stars; the most recent commit was on 11 November 2026.

Cross-client differences are easy to underestimate: a tool description that works well in one client can be ignored or misused in another. Running the same eval cases across clients before release is the most direct way to find those gaps.

Its OAuth debugger alone can save days, because authorisation failures are often opaque on the client side and depend on which version of the spec each client implements.

Because it can run locally against stdio servers, MCPJam also fits early development before a server is deployed anywhere.

Features

1. Interactive inspector: Tools, resources, prompts and elicitation.
2. JSON-RPC tracing: Every message and OAuth exchange.
3. Cross-client evals: 16 client configurations and 170+ models.
4. OAuth conformance: All MCP OAuth spec versions, DCR and CIMD.
5. Widget emulator: OpenAI Apps SDK and MCP app UIs.
6. Skills playground: Local and hosted skills.
7. CLI and SDK: Test in terminal or code.
8. CI/CD actions: Gate PRs on model behaviour.

Pros

1. Behavioural testing: Goes beyond protocol checks.
2. Free for developers: Test servers as you build.
3. Free to use: Open source under the Apache-2.0 (with its eval services under a separate licence) licence, so the core tool costs nothing to use.
4. Actively maintained: The repository had commits as recent as 11 November 2026.

Cons

1. Mixed licensing: Eval services under a separate licence.
2. Model costs: Evals against real LLMs cost tokens.

Usage and Links (Example)

npx @mcpjam/inspector@latest

Connect a local stdio or HTTP server, run tools in the playground, then add eval test cases and the CI action to gate pull requests.

Resources:

• MCPJam Inspector website
• GitHub repository

Pricing

MCPJam’s inspector is open source and free for developers testing servers as they build; hosted team and enterprise features are paid (see mcpjam.com).

Recommended for?

1. MCP server and ChatGPT app developers.
2. Teams shipping MCP integrations to many clients.

Conclusion

MCPJam is the most complete MCP testing platform, covering protocol, OAuth, widgets and model behaviour.

Start locally, then add cross-client evals to CI.

On CheckOps, MCPJam Inspector is filed under the API & contract testing, Integration testing and Release & CI gates stages. Related tools listed here include Promptfoo, Playwright MCP, Chrome DevTools MCP, Postman, DeepEval and MCP Selenium. When comparing them, weigh the licence, the languages and platforms each supports, whether it is self-hosted or managed, and how cleanly it slots into your existing CI pipeline. Details on this page were checked against the official website, documentation and repository on 11 November 2026.