MCP Test Servers

A collection of test servers implementing the Model Context Protocol (MCP).

🚀 Quick Install Gallery

Visit our GitHub Pages Gallery for one-click install buttons that work directly with Cursor!

Manual Installation

npx -y @msfeldstein/mcp-test-servers <server>

Available Servers

• all-types: Demonstrates various tool parameter types supported by MCP
• bad-param: Server with an intentionally malformed parameter name
• big-response: Server that returns large responses
• broken-tool: Server with intentionally broken tool
• broken-schema: Server with invalid JSON schema using oneOf/allOf/anyOf at top level
• broken-schema-fastmcp: FastMCP server with genuinely broken schemas using oneOf/allOf/anyOf at top level
• crash-on-startup: Server that crashes on startup
• combined: Server with tools and resources
• duplicate-names: Server with duplicate names for resources
• elicitation: Server demonstrating MCP elicitation feature - requests user information (name, age, contact details, preferences) through structured forms
• enum-param: Tool has enum string parameter
• env-check: Checks for SHOULD_RUN environment variable being passed properly
• env-echo: Echoes the environment variables
• gpt5-breaker: Server with SQL query execution tool using complex parameter validation
• headers: FastMCP server that safely returns HTTP header information
• instructions: Server with instructions and a makeRequest tool that returns success for value 42
• image: Tool returns an image of sonic the hedgehog
• long-description: Publicize a very long description configured via env var
• long-running: Server that sends progress notifications every 2 seconds for a 20-second task
• many-resources: Server with multiple resources
• many-tools: Server with 100 tools that each return 'ack'
• math: Server with basic math function tools (add, subtract, multiply, divide, power, sqrt, factorial)
• missing-type: Server with tools missing the required 'type' field in parameter schemas
• missing-type-fastmcp: FastMCP server with genuinely missing 'type' fields in tool schemas
• named: Server with configurable name via MCP_SERVER_NAME environment variable
• notion-gemini-repro: Server with complex nested schema that breaks on some models (Notion API reproduction case) - inline schema
• notion-gemini-repro-raw: Server with complex nested schema that breaks on some models (Notion API reproduction case) - exact raw schema object
• number-param: Tool with a number parameter
• oauth-repro: Reproduces OAuth 2.0 flow issues with proper discovery endpoints - runs resource server on port 3001 and auth server on port 3002
• oauth-debug: OAuth debugging server with comprehensive logging and endpoint tracking - runs resource server on port 3003 and auth server on port 3004
• oauth-callback-test: OAuth callback testing server that simulates Cursor 1.4.x endpoint discovery bugs - runs on port 3005
• oauth-edge-case: OAuth server with configurable edge cases to trigger discovery bugs - resource server on port 3006, auth server on port 3007
• oauth-bug-trigger: Specialized server designed to reproduce the exact Cursor 1.4.x OAuth discovery bug - resource server on port 3008, auth server on port 3009
• optional-param: Tool has an optional param
• pattern-param: Tool has a parameter with a pattern match
• ping: A simple server that responds with 'pong'
• prompts: Server with two prompts - one static greeting and one dynamic story generator with character name and location parameters
• raw-broken: Raw JSON-RPC MCP server with genuinely broken schemas (no framework validation)
• resource: Resource server implementation
• resource-lister: Server that publishes 5 resources and has a tool to list them
• roots-echo: Server that demonstrates MCP roots functionality by echoing back the roots provided by the client
• session-inactivity: Streamable HTTP server that expires sessions after 30 seconds of inactivity to test session management
• session-management: HTTP server that implements MCP session management behavior - assigns session IDs, requires them in requests, returns 404 for terminated sessions
• sse-timeout: Streamable HTTP server that establishes SSE connections but times out after 30 seconds to test client reconnection
• stderr: Server that logs to stderr
• structured-output: Demonstrates structuredContent responses with a simple echo tool
• dual-content: Demonstrates different outputs for structuredContent and content in the same response
• everything: Comprehensive server testing all MCP features - tools with/without params, resources, prompts, elicitation, dynamic tools, roots echo, env vars

🚀 Quick Start: Automated Testing

For instant comprehensive testing, use the built-in AI agent testing prompt:

1. Start the everything server: npx @msfeldstein/mcp-test-servers everything
2. In Cursor, use the slash command: /test-everything
3. The AI will automatically execute all 20 test steps and provide a complete validation report

Test Plan for Everything Server

The everything server is designed to comprehensively test all MCP features. Here's how to test each feature:

Tools Testing

0. Setup

{
    "mcpServers": {
        "everything": {
            "command": "npx @msfeldstein/mcp-test-servers everything",
            "env": {
                "MY_ENV_VAR": "My Special Token"
            }
        }
    }
}

1. Tool with Parameters

Tool: echo_with_params Test: Call with various parameter combinations:

- message: "Hello World", repeat_count: 3, uppercase: true
- message: "Test", repeat_count: 1, uppercase: false
- message: "MCP" (using defaults)

Expected: Should echo the message with specified repetitions and case transformation.

2. Tool without Parameters

Tool: simple_ping Test: Call without any parameters Expected: Should return "pong"

3. Tool Returning Mixed Content

Tool: get_mixed_resources Test: Call the tool Expected: Should return both text content and a small test image in the same response

4. Elicitation Testing

Tool: test_all_elicitations Test: Call the tool and respond to all prompts:

• String input (name)
• Boolean input (programming preference)
• Number input (years of experience, 0-50)
• Enum input (programming language selection) Expected: Should collect all inputs and display a summary

5. Dynamic Tool Management

Tool: toggle_dynamic_tool Test:

1. Call toggle_dynamic_tool to enable the dynamic tool
2. Verify dynamic_feature tool becomes available
3. Call dynamic_feature with optional action parameter
4. Call toggle_dynamic_tool again to disable
5. Verify dynamic_feature tool is no longer available Expected: Tool availability should change dynamically

6. MCP Roots Echo

Tool: echo_mcp_roots Test: Call the tool Expected: Should return JSON representation of MCP roots provided by the client

7. Environment Variable Echo

Tool: echo_env_var Test: prompt: "Call the echo_env_var MCP tool" Expected: Should return "MY_ENV_VAR = My Special Token"

8. Long-Running Progress Tool

Tool: long_running_progress Test: Call with optional task_name parameter Expected: Should run for 20 seconds with progress updates every 2 seconds, showing:

• Progress from 0 to 10 (total steps)
• Descriptive messages for each step
• Final completion message
• Client should receive progress notifications

Resources Testing

9. Text Resource

Resource: everything://docs.txt Test: Request the resource Expected: Should return comprehensive documentation about the server

10. Image Resource

Resource: everything://test-image.png Test: Request the resource Expected: Should return a small test image in PNG format

Prompts Testing

11. Prompt without Parameters

Prompt: simple-greeting Test: Type /sim and /simple-greeting should be suggested. Submit it. Expected: Should return a friendly greeting message about the Everything Server

12. Prompt with Parameters

Prompt: personalized-message Test: submit /personalized-message:

• name: "Alice"
• topic: "AI Development" Expected: Should return a personalized message mentioning Alice and AI Development

13. AI Agent Testing Prompt

Prompt: test-everything Test: Use this prompt in Cursor with a slash command: /test-everything Expected: Should return comprehensive step-by-step testing instructions for an AI agent to systematically test all MCP features Usage: Perfect for automated testing - the AI will execute all 20 testing steps automatically and provide a complete validation report

Complete Test Sequence

1. Setup: Set MY_ENV_VAR="My Special Token" in environment
2. Basic Tools: Test simple_ping and `echo_with_params``
3. Mixed Content: Test get_mixed_resources
4. Elicitation: Run through test_all_elicitations completely
5. Dynamic Tools: Toggle dynamic tool on/off and test availability
6. System Info: Test echo_mcp_roots and echo_env_var
7. Long-Running: Test long_running_progress with progress notifications
8. Resources: Fetch both text and image resources
9. Prompts: Test both parameterized and non-parameterized prompts

This comprehensive test ensures all MCP protocol features are working correctly.