MCPJam Inspector: Test and Debug MCP Servers Locally

TL;DR

TL;DR: MCPJam Inspector is a local development client for testing MCP servers and ChatGPT apps — chat with AI models, trace every JSON-RPC call, validate OAuth flows, and run evals in CI/CD, all from your terminal or desktop.

What Is MCPJam Inspector?

MCPJam Inspector is an open-source development platform for testing and debugging MCP (Model Context Protocol) servers and ChatGPT apps. It runs as a hosted web app, a desktop client for macOS and Windows, or directly in your terminal via npx.

The core workflow: point Inspector at your MCP server URL (or connect via STDIO locally), then chat with any of three AI models — GPT-5.5, Opus 4.7, or Gemini 3.1 — while Inspector captures every JSON-RPC message, tool call, argument, and response as a trace row. Latency and token counts are attached to each step.

The project is at github.com/MCPJam/inspector with 2025 stars and an Apache 2.0 license.

Source and Accuracy Notes

• Project page: mcpjam.com
• Source repository: github.com/MCPJam/inspector
• License: Apache 2.0 (verified via LICENSE file in repo)
• HN launch thread: news.ycombinator.com/item?id=46815786
• npm package: @mcpjam/inspector (v2.17.0)
• Source last checked: 2026-06-22

Setup

Option 1: Web App (no install)

Open app.mcpjam.com in your browser. No install needed, always on the latest version. Share server links with teammates the same way you’d share a Google Doc.

Note: the hosted app supports HTTPS MCP server URLs only. For HTTP or local STDIO servers, use the Desktop or Terminal options.

Option 2: Terminal

Requires Node.js 20+. Run:

npx @mcpjam/inspector@latest

After startup, Inspector prints a localhost URL to open in your browser. The terminal runner supports both HTTP/S and local STDIO servers.

Option 3: Desktop App

Download the installer for your OS. No Node.js required.

• Mac (DMG)
• Windows (EXE)

Key Features

Chat with full trace visibility

Send a prompt to an AI model and see exactly which tool it picks, what arguments it passes, and what your server returns — for every turn in the conversation. Three models are available in one view: GPT-5.5, Opus 4.7, and Gemini 3.1, with per-model latency and token usage displayed.

Protocol conformance testing

Inspector can validate your server against multiple MCP spec versions: 11-05-24, 3-26-25, 6-18-25, and 11-25-25. It also tests all client registration methods including Dynamic Client Registration (DCR), pre-registration, and Client ID Metadata (CIMD).

OAuth debugger

Step through every redirect and token exchange in your auth flow with a protocol-level trace. Useful for catching mismatches between what your server expects and what the client sends.

Local-to-production tunnel

Test a local server running behind a firewall with built-in ngrok integration. Run prompts inside actual clients like ChatGPT or Claude while viewing full logs in Inspector.

UI preview

Preview how your MCP tools render in ChatGPT or Claude’s UI, with mobile/tablet/desktop viewport toggles and light/dark mode switching.

CLI for terminal workflows

Probe servers, run doctor checks, exercise OAuth flows, and list tools/resources/prompts directly from the terminal:

# List all tools, resources, and prompts from a server
npx @mcpjam/inspector@latest list --url https://your-server.com/mcp

# Run a doctor check against your server
npx @mcpjam/inspector@latest doctor --url https://your-server.com/mcp

SDK and CI/CD integration

Programmatically drive inspections from your own test suite using the SDK. Wire the CLI into GitHub Actions to run end-to-end tests, evals, OAuth checks, and spec conformance on every PR:

# In your GitHub Actions workflow
npx @mcpjam/inspector@latest eval --url ${{ env.MCP_SERVER_URL }} --model gpt-5.5

Practical Evaluation Checklist

Use this checklist when evaluating an MCP server with Inspector:

1. Connect — Point Inspector at your server URL or start a local STDIO session.
2. Inspect — Use the Inspector tab to explore your server’s declared tools, resources, and prompts. Verify shapes match your expectations.
3. Chat — Send a prompt that should trigger a specific tool. Confirm the right tool is called with the right arguments.
4. Trace — Review the full trace table. Each row shows: agent step, tool call, arguments, server response, latency, token count.
5. Conformance — Run the spec conformance check. Fix any protocol deviations before shipping.
6. OAuth — If your server uses OAuth, step through the flow in the Auth tab. Confirm redirect URIs, scopes, and token exchanges are correct.
7. Eval — Run an eval across multiple LLMs. Track accuracy over time to catch regressions in future changes.

Security Notes

• Inspector captures all JSON-RPC traffic between the AI model and your server. Do not connect to untrusted or third-party MCP servers in Inspector — treat it like a network proxy.
• The OAuth debugger logs redirect URIs and token exchange parameters. Clear your session logs after testing production auth flows.
• The hosted web app at app.mcpjam.com is HTTPS-only. Local STDIO connections never leave your machine.
• Desktop and terminal versions support both HTTP and STDIO transports. Prefer STDIO for servers running on the same host to avoid exposing credentials over the network.

FAQ

Q: Do I need a ChatGPT or Claude subscription to use Inspector? A: No. Inspector provides its own AI chat interface with GPT-5.5, Opus 4.7, and Gemini 3.1. You connect your own API keys or use the models’ default billing. Your existing ChatGPT/Claude subscription is not required.

Q: What’s the difference between the hosted web app and the desktop/terminal versions? A: The hosted app at app.mcpjam.com requires no install but only supports HTTPS server URLs. The desktop and terminal versions additionally support HTTP servers and local STDIO connections, plus the tunnel, skills, and tasks features.

Q: Can I use Inspector in automated tests? A: Yes. The SDK lets you drive inspections programmatically from your test suite, and the CLI integrates into any CI/CD pipeline. The GitHub Actions workflow example in the README shows how to run evals on every PR.

Q: Does Inspector support MCP spec versions older than the current one? A: Inspector validates against multiple spec versions (11-05-24, 3-26-25, 6-18-25, 11-25-25) so you can test compatibility with clients still on older spec versions.

Q: How does Inspector compare to manually testing with ChatGPT or Claude? A: Manual testing requires deploying your server and switching between the chat UI and your logs. Inspector keeps everything in one view — chat, trace, and server logs — and lets you replay the same prompt across three models simultaneously with per-turn latency and token breakdowns.

Conclusion

MCPJam Inspector fills the gap between “it works in my head” and “it works in production” for MCP server developers. The trace view alone — seeing every JSON-RPC call, argument, and response with latency attached — is worth the setup. The OAuth debugger, spec conformance checker, and CI/CD integration round out a tool that covers the full development lifecycle from first local test to automated regression checks on every commit.

If you’re building MCP servers or integrating MCP clients, Inspector is worth adding to your dev toolbox. The hosted web app needs zero install to get started.