MCPJam Inspector - Local Development and Testing Tool for MCP Servers

Overview

MCPJam Inspector is a comprehensive local development and testing tool for MCP servers, ChatGPT apps, and MCP Apps (ext-apps/SEP-1865). It provides a full widget emulator for ChatGPT and Claude app development, an LLM playground with free access to frontier models (GPT-5, Claude Sonnet), OAuth debugging, test case management with AI generation, and E2E evaluation capabilities via CLI.

Core Value Proposition: Eliminates the need for ngrok or ChatGPT subscriptions during MCP server and app development by providing a complete local testing environment with multi-LLM support.

Problem Addressed

Key Statistics (as of January 26, 2026)

Key Features

1. MCP Server Inspection

• Multi-Server Connection: Connect to multiple MCP servers simultaneously
• Tool Testing: Manually invoke tools with custom parameters, view JSON-RPC logs
• Resource & Prompt Testing: Test resources, resource templates, and prompts
• Server Information: View icons, versions, capabilities, and instructions
• All JSON-RPC Logging: Real-time protocol message inspection

2. ChatGPT Apps SDK Support

• Full window.openai API: Complete support for widgetState, callTool, structuredContent, sendFollowUpMessage, displayMode
• Widget Emulator: Preview widgets without ChatGPT subscription
• Device Emulation: Test across Desktop, Tablet, Mobile viewports
• CSP Testing: Validate Content Security Policy configurations
• Theme/Locale Testing: Light/dark modes, internationalization testing
• Safe Area Insets: Simulate device notches and gesture areas

3. MCP Apps (SEP-1865) Support

• Full JSON-RPC Message Types: tools/call, ui/initialize, ui/message, ui/open-link
• Inline Widget Rendering: Render ui.resourceUri components in chat
• Display Modes: Inline, Picture-in-Picture, Fullscreen support

4. OAuth Debugger

• Visual Step-by-Step: Interactive OAuth handshake walkthrough
• Multi-Protocol Versions: Supports specs 03-26, 06-18, and 11-25
• Registration Methods: Client pre-registration, DCR, and CIMD
• Network Inspection: View all HTTP requests/responses with headers
• Sequence Diagrams: Visual flow synchronized with progress

5. LLM Playground

• Free Frontier Models: GPT-5, Claude Sonnet, Gemini 2.5 at no cost
• Multi-Provider Support: OpenAI, Anthropic, Google, Deepseek, Mistral, Ollama, OpenRouter, LiteLLM
• Split-Panel Interface: Chat on left, JSON-RPC logs on right
• MCP Prompts: Use prompts directly via / command
• Elicitation Support: Handle interactive prompts from MCP servers
• Customizable Agent: System prompts, temperature settings

6. Test Cases

• AI-Powered Generation: Auto-generate tests from tool definitions using Claude Haiku
• Positive/Negative Tests: Test both tool triggering and non-triggering scenarios
• Multi-Model Evaluation: Run tests across different LLM providers
• Accuracy Metrics: Track pass rates, token consumption, duration
• Run History: Compare performance across runs and models
• App Store Compliance: Generate required test cases for ChatGPT app submission

7. MCP Evals (CLI)

• E2E Testing: Simulate end-user environments
• Environment Simulation: Test as Claude Desktop, Cursor, etc.
• Multiple Servers: Test against multiple MCP servers
• CI/CD Integration: JSON-based configuration for automation

Technical Architecture

Deployment Options

npx (Quick Start):

npx @mcpjam/inspector@latest

Docker:

docker run -p 127.0.0.1:6274:6274 mcpjam/mcp-inspector

Desktop App: Native Mac (.dmg) and Windows (.exe) installers available

With MCP Server:

npx @mcpjam/inspector@latest npx -y @modelcontextprotocol/server-everything

Configuration

Supports Claude Desktop-style configuration files:

{
  "mcpServers": {
    "weather-server": {
      "command": "python",
      "args": ["weather_server.py"],
      "env": {
        "WEATHER_API_KEY": "${WEATHER_API_KEY}"
      }
    }
  }
}

Launch with: npx @mcpjam/inspector@latest --config path/to/config.json

Installation and Usage

Quick Start

# Start inspector
npx @mcpjam/inspector@latest

# Open browser to http://localhost:6274

Start with MCP Server

# Python FastMCP server
npx @mcpjam/inspector@latest uv run fastmcp run /path/to/server.py

# Node.js server
npx @mcpjam/inspector@latest npx -y your-mcp-package

Start with Ollama

npx @mcpjam/inspector@latest --ollama llama3.2

MCP Evals CLI

# Install CLI globally
npm install -g @mcpjam/cli

# Run evaluations
mcpjam evals run --tests tests.json --environment env.json

Test File Format

{
  "tests": [
    {
      "title": "Test weather tool",
      "query": "What's the weather in San Francisco?",
      "expectedTools": ["get_weather"],
      "model": { "id": "claude-3-5-sonnet-20241022", "provider": "anthropic" },
      "selectedServers": ["weather-server"]
    }
  ]
}

Comparison with Alternatives

Relevance to Claude Code Development

Direct Applications

1. MCP Server Development: Test custom MCP servers before integrating with Claude Code
2. OAuth Flow Debugging: Debug authentication for MCP servers requiring OAuth
3. Tool Testing: Manually invoke and verify tool behavior before deployment
4. Multi-LLM Validation: Verify MCP servers work across Claude, GPT, Gemini models

Patterns Worth Adopting

1. Visual OAuth Debugging: Step-by-step handshake visualization pattern
2. AI Test Generation: Using LLMs to generate test cases from tool definitions
3. E2E Eval Framework: JSON-based test configuration for CI/CD integration
4. Free Frontier Model Access: Proxied access pattern for development tools
5. Widget Emulation: Iframe isolation with injected APIs for UI testing
6. Multi-Protocol Support: Backwards-compatible OAuth spec handling

Integration Opportunities

1. Claude Code Plugin: Create plugin that launches MCPJam for connected MCP servers
2. Pre-Deployment Testing: Integrate MCPJam evals into plugin CI/CD pipelines
3. OAuth Debugging Workflow: Use MCPJam for debugging MCP OAuth implementations
4. Test Case Library: Generate and maintain test cases for Claude Code MCP integrations

Complementary to Existing MCP Ecosystem Research

MCPJam fills a testing and development tooling gap:

• Narsil MCP: Code intelligence and security scanning (runtime)
• OctoCode MCP: Research Driven Development with GitHub search (research)
• Docs MCP Server: Documentation grounding and retrieval (knowledge)
• MCPJam: Development, testing, and evaluation (DevOps/QA)

References

1. Documentation: https://docs.mcpjam.com (accessed 2026-01-26)
2. GitHub Repository: https://github.com/MCPJam/inspector (accessed 2026-01-26)
3. npm Package (@mcpjam/inspector): https://www.npmjs.com/package/@mcpjam/inspector (accessed 2026-01-26)
4. npm Package (@mcpjam/cli): https://www.npmjs.com/package/@mcpjam/cli (accessed 2026-01-26)
5. Getting Started Guide: https://docs.mcpjam.com/getting-started.md (accessed 2026-01-26)
6. OAuth Debugger: https://docs.mcpjam.com/inspector/guided-oauth.md (accessed 2026-01-26)
7. App Builder: https://docs.mcpjam.com/inspector/app-builder.md (accessed 2026-01-26)
8. LLM Playground: https://docs.mcpjam.com/inspector/llm-playground.md (accessed 2026-01-26)
9. Test Cases: https://docs.mcpjam.com/inspector/test-cases.md (accessed 2026-01-26)
10. MCP Evals: https://docs.mcpjam.com/evals/overview.md (accessed 2026-01-26)
11. MCP Protocol Specification: https://modelcontextprotocol.io/ (accessed 2026-01-26)
12. ChatGPT Apps SDK: https://developers.openai.com/apps-sdk/ (accessed 2026-01-26)
13. MCP Apps SEP-1865: https://github.com/modelcontextprotocol/modelcontextprotocol/pull/1865 (accessed 2026-01-26)