DEV Community

TheAIRabbit
TheAIRabbit

Posted on

Testing MCP Servers like a Pro using MCPJam Inspector

#ai #mcp #testing #tooling

Building and testing MCP (Model Context Protocol) servers is frustrating without the right tools. Most developers waste hours switching between different environments—writing code, then switching to clients like Cursor or Claude Desktop just to test a simple function call, then back to the IDE to debug issues. You're constantly guessing what's wrong when tools fail: Is it the MCP protocol implementation? The connection parameters? The tool definition? MCPJam Inspector solves this by giving you a dedicated, visual workspace for testing, debugging, and validating MCP servers without ever leaving your development flow. It's the difference between fumbling in the dark and having X-ray vision into your MCP implementation.

What is MCPJam Inspector?

MCPJam Inspector is a local-first developer tool for testing, debugging, and inspecting Model Context Protocol (MCP) servers and ChatGPT/OpenAI apps. Think of it as "Postman for MCP"—a visual interface that lets you explore, test, and debug MCP servers without needing to deploy them or connect through production clients.

Key Features

  • Visual Server Management: Connect to MCP servers via STDIO, HTTP, or SSE protocols
  • Tool Testing: Manually invoke and test MCP tools with custom parameters
  • Resource Inspection: Browse and fetch resources exposed by MCP servers
  • Prompt Templates: Test and use prompt templates with slash commands
  • LLM Playground: Simulate how your MCP server performs with various LLMs (OpenAI, Claude, Ollama, etc.)
  • Real-time Logging: View all JSON-RPC messages, requests, responses, and errors
  • OAuth Debugging: Test and debug OAuth flows for authenticated servers
  • Chat Interface: Interact with your MCP server conversationally

What Can You Use It For?

  • Development: Build and test MCP-based tools locally without switching to clients like Cursor or Claude Desktop
  • QA & Debugging: Validate tool definitions, prompt templates, and resource endpoints against the MCP specification
  • Experimentation: Test your MCP server with different LLM models to see how it behaves in various contexts
  • Learning: Understand how MCP servers work by inspecting the protocol messages in real-time
  • Integration Testing: Verify that your MCP server works correctly before deploying to production

About This Tutorial

This tutorial demonstrates how to use MCPJam Inspector to add and test MCP servers. We use the Tavily MCP server as an example, but the same process works for any MCP server:

  • Custom MCP servers you've built
  • Third-party MCP servers (GitHub, Slack, Notion, etc.)
  • Local MCP servers running on your machine
  • Remote MCP servers via HTTP/SSE

The steps are identical—just replace the server URL and configuration with your own MCP server details.

Overview

This tutorial walks you through using MCPJam Inspector to:

  1. Add an MCP server (using Tavily as an example)
  2. Connect via HTTP/SSE
  3. View available tools from the server
  4. Test the tools with custom parameters

Prerequisites

  • MCPJam Inspector running at http://127.0.0.1:6274
  • An MCP server to connect to (we'll use Tavily as an example - get an API key from Tavily's website if following along)

Step 1: Open MCPJam Inspector

Navigate to http://127.0.0.1:6274 in your browser. You'll see the main dashboard with no servers connected.

Initial State

Step 2: Click "Add Server"

Click the "Add Server" button in the top right corner of the MCP Servers section.

Add Server Dialog

Step 3: Select HTTP Connection Type

The dialog opens with STDIO selected by default. Click the Connection Type dropdown and select "HTTP".

Connection Type Dropdown

After selecting HTTP, the form changes to show HTTP-specific fields:

  • Server Name: Enter a name for your server
  • URL: Enter the Tavily MCP server URL
  • Authentication: Configure if needed
  • Custom Headers: Add any custom headers

HTTP Connection Form

  1. Server Name: Enter a name for your server (we used tavily as an example)
  2. URL: Enter your MCP server URL. For the Tavily example: 
   https://mcp.tavily.com/mcp/?tavilyApiKey=YOUR_API_KEY
Enter fullscreen mode Exit fullscreen mode

Note: Replace YOUR_API_KEY with your actual API key. For other MCP servers, use their respective connection URLs.

Server Name Filled

URL Filled

Click the "Add Server" button at the bottom of the dialog. The server will connect automatically.

Server Connected

You can see:

  • Server name: tavily (or whatever you named it)
  • Connection type: HTTP/SSE
  • Status: Connected
  • Server version: v2.14.2

Click on "Tools" in the left sidebar to see all available tools from your MCP server.

Tools List

In our example with Tavily, we see 4 tools:

  1. tavily_search - Search the web for real-time information
  2. tavily_extract - Extract content from specific web pages
  3. tavily_crawl - Crawl multiple pages from a website
  4. tavily_map - Map and discover website structure

Different MCP servers will expose different tools based on their functionality.

Step 8: Test a Tool

Click on any tool from your MCP server to open its configuration form. In our example, we'll test "tavily_search".

Tool Form

The form shows all available parameters for the selected tool. Each MCP server's tools will have different parameters based on their functionality.

Step 9: Enter Parameters

Fill in the required parameters. For the tavily_search example, enter a test query like: MCP protocol tutorial

Query Filled

Step 10: Execute the Tool

Click the "Execute" button to run the tool. The button will show "Running" while processing.

Search Results

The results will appear in the Response section below, showing the tool's output in a structured format. The exact format depends on what the tool returns.

Next Steps

Explore More Tools

  • Try other tools from your MCP server
  • Test different parameter combinations
  • View the logs to see the JSON-RPC messages being exchanged

Use MCPJam Inspector's Advanced Features

  • Chat Interface: Interact with your MCP server conversationally using natural language
  • LLM Playground: Test how different LLMs (OpenAI, Claude, Ollama) use your MCP server's tools
  • Resource Templates: If available, explore prompt templates for standardized tool usage
  • Tracing: Monitor detailed request/response flows to understand how the MCP protocol works
  • Test Cases: Create and save test cases for automated testing of your MCP integrations

Happy Coding!

Top comments (1)

Collapse
 
kxbnb profile image
kxbnb •

Solid walkthrough of MCPJam Inspector! The focus on visibility into MCP server behavior is something more teams need to prioritize.

One pattern we've found useful is separating inspection from testing - being able to see exactly what requests the MCP server receives and what it returns, in real-time, without modifying the server code. At toran.sh we built a read-only inspection layer for this exact use case.

The challenge with most testing approaches is they require you to instrument the server. Sometimes you just want to "see what's happening" without changing anything - especially when debugging production issues or integrating with third-party MCP servers you don't control.

What's your experience with debugging MCP servers you don't have source access to?