name: mcp-testing description: > Unit and integration testing phase for MCP server development. Use when writing tests for MCP tool handlers, setting up a test harness, using a mock MCP client, testing error cases and edge inputs, validating tool output structure, or running automated tests against a locally running MCP server. Covers Vitest and pytest harness setup, mock client usage, handler unit tests, and Inspector-based integration testing.

MCP Testing

Unit tests for handlers and integration tests with a real transport.

Intent Router

Load reference files on demand — only when the corresponding topic is active:

• references/test-harness-setup.md — Load when setting up Vitest (TypeScript) or pytest (Python) for the first time, or configuring test fixtures
• references/mock-client-patterns.md — Load when writing protocol-level tests, testing JSON-RPC message flow, or simulating a client in-process

Quick Start — Test Pyramid

Start at the unit layer. Reach integration testing before mcp-verification. E2E testing belongs in mcp-test-drive.

Install test dependencies first:

npm install --save-dev vitest          # TypeScript
pip install pytest pytest-asyncio      # Python

TypeScript Unit Test Pattern

Separate handler logic from transport wiring to enable direct imports:

// src/tools/echo.ts
export async function echoHandler(args: { message: string }): Promise<string> {
  return args.message;
}

// tests/echo.test.ts
import { describe, it, expect } from "vitest";
import { echoHandler } from "../src/tools/echo.js";

describe("echoHandler", () => {
  it("returns message unchanged", async () => {
    expect(await echoHandler({ message: "hello" })).toBe("hello");
  });
  it("returns error string for invalid input", async () => {
    const result = await echoHandler({ message: "" });
    expect(typeof result).toBe("string");
  });
});

Run with: npm test (watch) or npm run test:run (single pass)

Python Unit Test Pattern

# tools/echo.py
def echo_handler(message: str) -> str:
    return message

# tests/test_echo.py
from tools.echo import echo_handler

def test_echo_returns_message():
    assert echo_handler("hello") == "hello"

def test_echo_empty_string():
    assert echo_handler("") == ""

Run with: pytest

MCP Inspector Integration

Run the Inspector to test the full server over a real transport:

# TypeScript (build first)
npm run build
npx @modelcontextprotocol/inspector node dist/index.js

# Python
npx @modelcontextprotocol/inspector python server.py

The Inspector opens a browser UI. Use the Tools panel to call each tool interactively and inspect raw JSON-RPC messages. Verify: all tools appear, all calls return valid content, error cases return isError: true.

Load references/mock-client-patterns.md for in-process integration tests that run without the Inspector browser UI.

Error Case Coverage

Every tool must be tested for these error cases before mcp-verification:

[ ] Missing required parameter → returns error content, not exception
[ ] Wrong parameter type → handled gracefully
[ ] Handler throws exception → caught and returned as isError: true
[ ] Resource not found → descriptive error message returned
[ ] Empty string parameter → handled without crash

Gate

mcp-testing is complete when:

• All unit tests pass (npm run test:run or pytest)
• Error case checklist above is fully covered
• Inspector shows all tools and at least one tool call returns a valid response

Proceed to mcp-verification only after unit tests pass and Inspector confirms tool registration.

Safety Notes

Never point tests at production data sources, live APIs, or shared databases. Use in-memory stubs, fixtures, and temporary directories for all test data. Tests that modify state should clean up in afterEach / teardown fixtures.

Resource Index