mcp-server-tester

MCP Server Tester

⚠️ WORK IN PROGRESS: This project is under active development and has not been thoroughly tested yet. Features may be incomplete, contain bugs, or change significantly. Use at your own risk in non-production environments only.

A powerful, configuration-driven testing tool for Model Context Protocol (MCP) servers. This project provides a comprehensive solution for validating, benchmarking, and ensuring reliability of MCP servers that integrate with AI models like Claude.

Current Status

This tool is moving toward an alpha release and currently offers:

• ✅ Basic configuration framework
• ✅ MCP server connection and CLI support
• ✅ Test generation using Claude AI
• ✅ Natural language query generation for tests
• ✅ Comprehensive response validation with multiple rules
• ✅ Report generation in console, JSON, HTML, and Markdown formats
• 🚧 Broader automated test coverage of the tester
• 🚧 Production hardening and packaging improvements

If you're interested in contributing, please feel free to open issues and submit pull requests.

Introduction

The Model Context Protocol (MCP) enables AI models to access external tools and data sources through standardized interfaces. As MCP servers grow in complexity and importance, ensuring their correct functionality becomes critical. The MCP Server Tester addresses this need by:

• Automating tests for all tools exposed by an MCP server
• Leveraging Claude AI to generate intelligent, contextually-relevant test cases
• Validating responses against expected outcomes and schemas
• Providing detailed reports to identify issues and performance bottlenecks

This tool is designed for MCP server developers, AI integration teams, and quality assurance professionals who need to ensure their MCP implementations are robust, reliable, and correctly follow the protocol specifications.

Concept

The Model Context Protocol is a standard that allows AI models to call external tools. An MCP server exposes one or more tools through a simple HTTP interface. Each tool describes its name, parameters, and response schema so that a model can invoke it safely.

mcp-server-tester automates the process of checking that an MCP server and its tools work correctly:

1. Discovery – it queries the server for all available tools.
2. Test generation – it uses Claude AI to create realistic test cases for each tool.
3. Execution – it runs those tests against the server.
4. Validation – it verifies the responses using configurable rules.
5. Reporting – it summarizes the results in the console or in JSON, HTML, or Markdown formats.

The goal is to quickly spot mismatches between expected and actual behaviour so that you can fix issues before exposing the tools to production models.

Purpose

• Reliability – catch bugs or inconsistent behaviour in your MCP server.
• Regression testing – run the same set of tests whenever the server changes.
• Documentation – generated reports describe the queries and expected outcomes for each tool.
• Automation – integrate the tester into CI pipelines to ensure ongoing quality.

Repository

• GitHub: https://github.com/r-huijts/mcp-server-tester
• Issues: https://github.com/r-huijts/mcp-server-tester/issues
• License: MIT

Features

• 🔍 Automatically discovers available tools from any MCP server
• 🧪 Generates realistic test cases for each tool using Claude AI
• ⚡ Executes tests and validates responses
• 📊 Provides detailed test reports
• 🔑 Supports multiple connection methods through configuration
• Configuration-Based: Simple JSON configuration for defining MCP servers to test
• Multiple Server Support: Test multiple MCP servers at once
• Comprehensive Testing: Tests all tools exposed by each server
• Natural Language Context: Includes the user query that would trigger each tool, providing real-world context
• Detailed Reports: Generate reports in console, JSON, HTML, or Markdown formats
• Secure: Keeps API keys in environment variables, not in configuration files

Prerequisites

• Node.js 18 or higher
• An Anthropic API key for generating test cases

Installation

Since this project is still in development, installation is done by cloning the repository:

# Clone the repository
git clone https://github.com/r-huijts/mcp-server-tester.git
cd mcp-server-tester

# Install dependencies
npm install

# Build the project
npm run build

# Create a symbolic link to use it globally (optional)
npm link

Configuration-Based Usage

The MCP Server Tester is designed to be driven entirely through configuration files. This approach offers several advantages:

• Reusability: Define your servers once, test them repeatedly
• Version control: Check in your test configurations alongside your code
• Sharing: Easily share server test configurations with team members

Basic Usage

# Create a .env file with your Anthropic API key
echo "ANTHROPIC_API_KEY=your-api-key-here" > .env

# Run tests using the configuration
mcp-server-tester

# Use a custom configuration file
mcp-server-tester path/to/my-config.json

Configuration File Structure

The configuration file (mcp-servers.json) controls all aspects of testing:

{
  "numTestsPerTool": 3,
  "timeoutMs": 10000,
  "outputFormat": "console",
  "outputPath": "./reports/results.json",
  "verbose": false,
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "./"],
      "env": {
        "DEBUG": "true"
      }
    },
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "your-github-token-here"
      }
    },
    "dev-server": {
      "command": "node",
      "args": ["/absolute/path/to/your/dev-server.js"],
      "env": {
        "DEBUG": "true",
        "NODE_ENV": "development"
      }
    }
  }
}

By default, the tool will test all servers defined in the mcpServers section. If you want to test only specific servers, you can add an optional servers array:

{
  "servers": ["filesystem", "dev-server"],
  "numTestsPerTool": 3,
  // other settings...
  "mcpServers": {
    // server definitions...
  }
}

Configuration Options

Test Settings

Option	Description	Default
servers	Optional array of specific server names to test	All servers in mcpServers
numTestsPerTool	Number of tests to generate per tool	3
timeoutMs	Timeout for test execution in milliseconds	10000
outputFormat	Format for test reports (json, console, html, markdown)	"console"
outputPath	Path to output file	undefined
verbose	Enable verbose logging	false

Server Definitions

The mcpServers section defines all available servers that can be tested:

Property	Description	Required
command	Executable or command to run	Yes
args	Array of command-line arguments	Yes
env	Environment variables to set	Yes

Server Connection Types

You can define various types of MCP servers in your configuration:

NPM Package

"npm-package": {
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-github"],
  "env": {}
}

Local Script with Relative Path

"python-script": {
  "command": "python",
  "args": ["./servers/custom_server.py"],
  "env": {
    "PORT": "8080"
  }
}

Local Script with Absolute Path

Useful for testing development versions of servers:

"dev-server": {
  "command": "node",
  "args": ["/absolute/path/to/your/dev-server.js"],
  "env": {
    "DEBUG": "true",
    "NODE_ENV": "development"
  }
}

Socket Connection

"remote-socket": {
  "command": "nc",
  "args": ["localhost", "3000"],
  "env": {}
}

API Key Management

For security reasons, your Anthropic API key should only be set in one of these ways:

1. Environment variable: ANTHROPIC_API_KEY=your-api-key
2. .env file in your project directory:

   ANTHROPIC_API_KEY=your-api-key
   

3. Optionally set the Claude model:

   CLAUDE_MODEL=claude-3-opus
   

   If not provided, the tool defaults to claude-3-7-sonnet-20250219.

Important: Never put your API key in the configuration file, as it may be committed to version control.

Command-Line Options

MCP Server Tester supports minimal command-line options:

Option	Description
--init or -i	Create a default configuration file
--list or -l	List all servers defined in your configuration
--help or -h	Display help information
--servers or -s	Comma-separated list of servers to test
[config-path]	Specify a custom configuration file path

The --servers option overrides the servers array in your configuration file.

Test Generation Process

The tool uses Claude AI to automatically generate appropriate test cases for each tool exposed by the MCP server:

1. It discovers all available tools from the server
2. For each tool, it analyzes:
   • Tool name and description
   • Required and optional parameters
   • Parameter types and constraints
3. Claude generates multiple test cases per tool:
   • Happy path tests with valid inputs
   • Edge case tests with boundary values
   • Error case tests with invalid inputs

Each test case includes:

• Description of what's being tested
• Input parameters
• Expected outcome criteria

Validation Rule Types

Validation rules are used to check the structure and content of a tool response.
The following rule types are supported:

• contains – ensure a string or array contains a given value
• matches – check equality or a regular expression match
• hasProperty – verify that a property exists
• equals – assert that a value exactly matches the expected value
• arrayLength – require an array to have a specific length
• custom – invoke a user-defined validation function

Test Execution and Validation

For each server specified in the configuration (or all servers if none specified):

1. The tool connects to the server
2. It discovers all available tools
3. It generates test cases for each tool
4. It executes each test case against the server
5. It validates the responses against expected outcomes
6. It generates a report of the results

Reporting Options

The tool can generate reports in multiple formats, controlled by the outputFormat configuration option:

Console Output (Default)

Displays test results directly in the terminal.

JSON Report

Creates a structured JSON file at the path specified in outputPath.

HTML Report

Generates an HTML report with visualizations at the path specified in outputPath.

Markdown Report

Creates a portable Markdown file at the path specified in outputPath.

Complete Examples

Basic Setup and Testing

1. Create a default configuration file:

   mcp-server-tester --init
   

2. Edit the mcp-servers.json file to add your own servers and settings

3. Create a .env file with your Anthropic API key:

   echo "ANTHROPIC_API_KEY=your-api-key-here" > .env
   

4. Run the tests:

   mcp-server-tester
   

Testing a Dev Version of Your Server

To test a development version of your MCP server:

1. Add a configuration for your development server with the absolute path:

{
  "mcpServers": {
    "my-dev-server": {
      "command": "node",
      "args": ["/path/to/your/project/dist/server.js"],
      "env": {
        "DEBUG": "true",
        "NODE_ENV": "development"
      }
    }
  }
}

2. Run the tests:

mcp-server-tester

Testing Multiple Different Configurations

You can maintain different configuration files for different testing scenarios:

# Create different config files for different environments
cp mcp-servers.json config-dev.json
cp mcp-servers.json config-prod.json

# Edit each file with appropriate settings

# Run tests with specific config
mcp-server-tester ./config-dev.json
mcp-server-tester ./config-prod.json

Troubleshooting

Connection Issues

If you're having trouble connecting to an MCP server:

1. Verify the server configuration in your mcp-servers.json file
2. Check if the server supports the MCP protocol
3. Try increasing the timeoutMs for slower servers
4. Enable verbose logging by setting verbose: true
5. Check server process startup with environment variable DEBUG=true

API Key Issues

If you encounter API key issues:

1. Verify your Anthropic API key is valid
2. Make sure the API key is correctly set in your environment or .env file
3. Check for any spaces or extra characters in your API key
4. Confirm that the .env file is in the correct location (project root)

Tool Execution Failures

If tool executions are failing:

1. Ensure your server implements the MCP protocol correctly
2. Check the server logs for errors
3. Verify the tool parameters are valid
4. Increase the timeout if the tool takes longer to execute

Node.js Deprecation Warnings

punycode Module Deprecation Warning

If you encounter this warning:

(node:71439) [DEP0040] DeprecationWarning: The `punycode` module is deprecated. Please use a userland alternative instead.

This is a harmless warning from Node.js about an internal module being deprecated. It doesn't affect the functionality of the MCP Server Tester. The warning comes from one of the dependencies and will be resolved in future updates.

Solutions:

1. Ignore the warning - It doesn't affect functionality
2. Suppress warnings - Run with the NODE_NO_WARNINGS=1 environment variable:

   NODE_NO_WARNINGS=1 mcp-server-tester
   

3. Use the npm scripts - The included npm scripts already suppress these warnings:

   npm start
   

Development

To set up the development environment:

# Clone the repository
git clone https://github.com/r-huijts/mcp-server-tester.git
cd mcp-server-tester

# Install dependencies (including dev dependencies)
npm install
# Run the Jest test suite
npm test
# Lint the code for style issues
npm run lint

# Create your .env file
cp .env.example .env
# Edit .env and add your API key

# Run the tool in development mode
npm run dev

Code Structure

The canonical test generator lives at src/test-generator/TestGenerator.ts.
Older files such as src/generator/TestGenerator.ts have been removed to avoid
confusion. All imports should reference the module under src/test-generator/.

Packaging and Distribution

The project can be published as an npm package for easier installation.

# Compile TypeScript sources
npm run build

# Create a tarball of the package
npm pack

# Publish to npm (requires npm credentials)
npm publish

Only the contents of the dist folder and essential documentation are included
in the published package. The build step runs automatically when npm publish
is executed.

License

This project is licensed under the MIT license.

Installation (Quick Start)

1. Install Node.js 18 or newer.
2. Clone the repository and install dependencies:

git clone https://github.com/r-huijts/mcp-server-tester.git
cd mcp-server-tester
npm install
npm run build

3. Optionally link the package globally so the mcp-server-tester command is available system-wide:

npm link

Configuration Overview

All behaviour is controlled by a JSON configuration file (by default mcp-servers.json). The configuration lists which MCP servers to test and defines options such as timeouts and report formats.

Create the file with --init or copy the provided example:

mcp-server-tester --init
# or
cp mcp-servers.json.example mcp-servers.json

Edit the file to add your servers. A minimal example looks like:

{
  "numTestsPerTool": 2,
  "timeoutMs": 10000,
  "outputFormat": "console",
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "./"],
      "env": { "DEBUG": "true" }
    }
  }
}

Place your Anthropic API key in a .env file or export it as ANTHROPIC_API_KEY:

ANTHROPIC_API_KEY=your-api-key

Running Tests

With the configuration and environment variables in place, run:

mcp-server-tester

Use mcp-server-tester path/to/config.json to specify a different configuration or --servers filesystem,github to test only certain servers.

Understanding Results

By default, results appear in the console. You can also generate structured reports:

• JSON – machine readable data for further processing.
• HTML – easy to browse summary with collapsible sections.
• Markdown – shareable format suitable for code reviews.

The report lists each tool, the natural language query used to invoke it, whether the response met the validation rules, and details on any failures.

Troubleshooting Basics

• Connection issues – verify the server paths and increase timeoutMs if needed.
• API key problems – make sure ANTHROPIC_API_KEY is set with no extra spaces.
• Tool failures – inspect server logs and ensure your implementation follows the MCP specification.