Testing MCP Servers with MCP Inspector

MCP Inspector is a free browser tool for testing and debugging MCP servers. See raw messages, inspect schemas, and understand how LLMs use your tools.

Magnifying glass inspecting MCP server data with debug and inspect code visuals on circuit background.

Using and developing MCP (Model Context Protocol) servers is becoming increasingly common for software engineers. Over the past few months, I’ve been experimenting with both consuming and building MCP servers. One thing became clear early on: testing or debugging an MCP server is surprisingly difficult without the right tools.

Most of the time, you interact with MCP servers indirectly through an LLM or AI client, which makes it challenging to see raw MCP messages, understand tool schemas, inspect resources, or debug server behavior.

To help with this, I want to share a great free tool that my colleagues and I at work have found incredibly useful: MCP Inspector.

What MCP Inspector Can Do

MCP Inspector runs directly through npx, with no installation or setup required. When launched, it opens in your browser (on port 6274 by default) and provides a complete environment for understanding and testing MCP servers. It also gives you a better understanding of how LLMs use MCP servers to enhance their capabilities and knowledge.

Here are a few key features of MCP Inspector:

  • MCP Server Support
    • Works with MCP servers installed in your project (npm, PyPI, uv, etc.).
    • Supports locally developed servers (great when running in watch mode).
    • Supports publicly available MCP servers, including those that require authentication.
  • Connection Pane
    • Choose the transport (typically stdio).
    • Provide a command and arguments for launching the server.
    • Set environment variables.
    • Save or load configurations using a servers.json file.
  • Resources Tab
    • Lists resources exposed by the MCP server.
    • Shows metadata such as MIME type, title, and description.
      (Note: not all servers expose resources)
  • Prompts Tab
    • Shows any prompt templates defined by the server.
    • Allows testing them interactively. (Note: not all servers expose prompts)
  • Tools Tab
    • Displays all callable tools offered by the MCP server
    • Shows the full schema, input definitions, and descriptions.
    • Validates that responses match the server’s declared schema.
  • Bonus: Dark Mode
    • Supports both light and dark UI themes (this is a big deal for some folks)

Getting Started: Basic Usage

MCP Inspector offers a wide range of configuration options, but getting started is easy. This guide assumes you already have Node.js 22 or later installed. (Node.js 24 is the LTS version at the time of writing.)

Run MCP Inspector

You can start MCP Inspector without any configuration to verify it runs correctly. Run the following command:

npx @modelcontextprotocol/inspector

MCP Inspector should open automatically at: http://localhost:6274, and you should see a screen that looks similar to the screenshot below:

A blank instance of MCP Inspector running in a browser
A blank instance of MCP Inspector running in a browser

MCP Inspector Interface

Once the inspector loads in your browser, you’ll see:

Left Panel — Connection Setup

  • Enter the command that starts your MCP server.
  • Add optional arguments or environment variables.
  • Choose between manually creating a server entry or loading from a config file.
  • Adjust settings like timeouts and authentication.

Main Area

  • History: Captures the raw MCP messages exchanged with the server.
  • Server Notifications: Shows logs sent from the MCP server to the client.
  • Tabs (once connected): Tools, Resources, Prompts, Auth, etc.

Example: Connecting MCP Inspector to the AWS Documentation MCP Server

The AWS Documentation MCP Server is a popular, publicly available server that returns AWS documentation, search results, and recommendations. It requires no authentication, making it a great first server to explore.

In real projects, your MCP server will typically be part of your codebase, and you’ll run it directly (e.g., npm run dev, uv run, or a compiled binary). MCP Inspector supports this just fine. However, for this example, we won’t install anything. Instead, we’ll let MCP Inspector launch the server directly using uvx.

Starting MCP Inspector with AWS Documentation MCP Server

Run the following command in your terminal:

npx @modelcontextprotocol/inspector \
  -e AWS_DOCUMENTATION_PARTITION=aws \
  uvx awslabs.aws-documentation-mcp-server@latest

Running this single command:

  • Launches MCP Inspector
  • Sets environment variablesAWS_DOCUMENTATION_PARTITION=aws tells the server to use global AWS documentation rather than GovCloud or China partitions
  • Spawns the AWS Documentation MCP Server using uvx
  • Automatically connects Inspector to the server

After clicking the "connect" button from the UI, you should see a screen that looks like the screenshot below (notice the green connection indicator in the left panel):

A screenshot of the MCP Inspector Connected to the AWS Documentation MCP Server
MCP Inspector Connected to the AWS Documentation MCP Server

Note: The AWS Documentation MCP Server exposes tools only, which is common for this API-type MCP servers. It does not define prompts or resources, so those tabs will remain empty.

Using A Tool

Now that we’re connected to the AWS Documentation MCP Server, we can inspect the tools it exposes to AI agents or MCP clients. Click the Tools tab, then click List Tools to load the available operations:

Listing of tools exposed by AWS Documentation MCP Server which includes 3 tools: read, search, and recommend
Listing of tools exposed by AWS Documentation MCP Server

Next, select the search_documentation tool. The right panel will update to show the full schema, expected inputs, and description that an LLM would receive when calling this tool. MCP Inspector even provides a “Copy input” button if you want to reuse the call.

Try running a test search using:

  • search_phrase: Lambda runtime support for Python
  • limit: 2

Then click Run tool:

Requesting the search tool to search for "Lambda runtime support for Python" with a limit of 2 results
Search tool input

MCP Inspector will show both the request sent to the server and the validated response. Helpful copy buttons let you quickly copy either for debugging, documentation, or testing:

A JSON Search response for a Lambda Runtime search request
Search tool response

Using this tool provides a good example of how to use MCP Inspector and how to start inspecting and debugging MCP servers.

Next Steps

Once you’re comfortable with the basics, here are a few ways to continue exploring MCP Inspector:

1. Try Other Public MCP Servers

Many companies and OSS projects publish MCP servers:

  • Filesystem servers
  • Documentation servers
  • Issue-tracking servers
  • Git providers
  • Knowledge base servers
  • Cloud platform servers

Each one exposes different resources, prompts, or tools. The Awesome MCP Servers Site is a great way to discover more MCP servers.

2. Connect Your Own MCP Server

If you’re building an MCP server locally (Node, Python, Rust, Go, etc.), point MCP Inspector. Instructions for doing this are in the MCP Inspector README.MD on GitHub.

3. Use Inspector as Part of Your Development Workflow

If you regularly develop MCP servers, running Inspector side-by-side with your dev environment is a great development tool because it lets you:

  • Inspect every message
  • Validate schemas
  • Debug tool execution
  • Test prompts interactively

Summary

MCP Inspector is a free, surprisingly powerful tool for testing and debugging MCP servers. It also gives developers an inside look at how LLMs actually use the tools we build. It’s become one of my go-to utilities, so I wanted to pass it along to anyone who might find it helpful.