Tester

GA npm version CI License: MIT

A testing and evaluation framework for Model Context Protocol (MCP) servers. Write deterministic Playwright tests against your MCP tools, or run data-driven eval datasets — including LLM-based evaluation of tool discoverability.

Playwright Tests

The mcp Playwright fixture connects to your MCP server (stdio or HTTP) and exposes a high-level API for calling tools and asserting responses. Custom matchers keep assertions readable.

import { test, expect } from '@gleanwork/mcp-server-tester/fixtures/mcp';

test('read_file returns file contents', async ({ mcp }) => {
  const result = await mcp.callTool('read_file', { path: '/tmp/test.txt' });
  expect(result).toContainToolText('Hello, world');
  expect(result).not.toBeToolError();
});

test('server exposes required tools', async ({ mcp }) => {
  const tools = await mcp.listTools();
  expect(tools.map((t) => t.name)).toContain('read_file');
});

Playwright tests are fast, deterministic, and designed for CI. Use them for regression testing, schema validation, and protocol conformance. The framework includes built-in conformance checks for the MCP spec.

Available matchers:

Matcher	Description
toMatchToolResponse	Response exactly matches expected value (deep equal)
toContainToolText	Response contains expected substrings
toMatchToolSchema	Response validates against a Zod schema
toMatchToolPattern	Response matches a regex pattern
toMatchToolSnapshot	Response matches a saved baseline
toBeToolError	Response is (or is not) an error
toHaveToolResponseSize	Response size is within bounds
toSatisfyToolPredicate	Response satisfies a custom function
toHaveToolCalls	LLM called the expected tools
toHaveToolCallCount	LLM made N tool calls
toPassToolJudge	LLM evaluates response quality against a rubric

Eval Datasets

Eval datasets let you define test cases as JSON files and run them with runEvalDataset(). Each case specifies a tool call and one or more assertions.

{
  "name": "file-ops",
  "cases": [
    {
      "id": "read-config",
      "toolName": "read_file",
      "args": { "path": "/tmp/config.json" },
      "expect": {
        "schema": "file-content",
        "containsText": ["version", "name"]
      }
    },
    {
      "id": "read-readme",
      "toolName": "read_file",
      "args": { "path": "/tmp/README.md" },
      "expect": {
        "snapshot": "readme-snapshot"
      }
    }
  ]
}

import { test, expect } from '@gleanwork/mcp-server-tester/fixtures/mcp';
import { loadEvalDataset, runEvalDataset } from '@gleanwork/mcp-server-tester';
import { z } from 'zod';

test('file operations eval', async ({ mcp }, testInfo) => {
  const dataset = await loadEvalDataset('./data/evals.json', {
    schemas: { 'file-content': z.object({ content: z.string() }) },
  });
  const result = await runEvalDataset({ dataset }, { mcp, testInfo });
  expect(result.passed).toBe(result.total);
});

Supported assertion types:

Type	Description
containsText	Response includes expected substrings
schema	Response validates against a Zod schema
regex	Response matches a pattern
snapshot	Response matches a saved baseline
judge	LLM evaluates response quality against a rubric
toolsTriggered	LLM called the expected tools (LLM host mode)

LLM host mode

In LLM host mode, a real LLM receives your server's tool list and a natural language prompt, then decides which tools to call. This tests whether your tool names, descriptions, and input schemas are clear enough for autonomous use — a different question from whether the tools return correct output.

{
  "id": "find-config",
  "mode": "mcp_host",
  "scenario": "Find the application config file and return its contents",
  "mcpHostConfig": {
    "provider": "anthropic",
    "model": "claude-opus-4-20250514"
  },
  "expect": {
    "toolsTriggered": {
      "calls": [{ "name": "read_file", "required": true }]
    }
  }
}

LLM host mode makes real API calls and produces non-deterministic results. Use iterations to run a case multiple times and measure pass rate rather than expecting 100% on a single run. See the LLM Host Guide for configuration and cost management.

Installation

Requires Node.js 22+.

npm install --save-dev @gleanwork/mcp-server-tester @playwright/test

The Anthropic SDK is only needed for LLM-as-judge assertions or LLM host mode with the Anthropic provider:

npm install --save-dev @anthropic-ai/sdk

Quick Start

npx mcp-server-tester init

The CLI wizard creates a playwright.config.ts, example tests, and a sample eval dataset configured for your server. See the CLI Guide for all options.

Configuration

Point the framework at your MCP server in playwright.config.ts:

import { defineConfig } from '@playwright/test';

export default defineConfig({
  testDir: './tests',
  reporter: [['list'], ['@gleanwork/mcp-server-tester/reporters/mcpReporter']],
  projects: [
    {
      name: 'my-server',
      use: {
        mcpConfig: {
          transport: 'stdio',
          command: 'node',
          args: ['server.js'],
        },
      },
    },
  ],
});

For HTTP servers, set transport: 'http' and serverUrl. For servers that require OAuth, see the Transports Guide and CLI Guide for authentication setup, including CI/CD token management.

Documentation

• Quick Start — detailed setup and configuration
• Expectations — all assertion types including snapshot sanitizers
• LLM Host Simulation — tool discoverability testing
• API Reference
• Transports — stdio and HTTP configuration, OAuth
• CLI Commands — init, generate, login, token
• UI Reporter — interactive web UI for test results
• Development — contributing and building
• Migration Guide (v0.12 → v1.0) — upgrading from pre-1.0 releases

AI Skills

Install AI skills to help your coding assistant generate tests, eval datasets, and MCP host evals:

npx skills add -g gleanwork/mcp-server-tester

This installs skills globally so they're available across all your projects. Four skills are included:

Skill	Description
mcp-tester-guide	Framework reference — matchers, config, auth, anti-patterns
write-mcp-test	Generate direct-mode Playwright tests
write-mcp-eval	Generate data-driven eval datasets
write-mcp-host-eval	Generate LLM host simulation evals

Compatible with Claude Code, Cursor, Windsurf, Copilot, and 40+ other AI agents.

Examples

The examples/ directory contains complete working examples:

• filesystem-server/ — Test suite for Anthropic's Filesystem MCP server: 5 Playwright tests, 11 eval dataset cases, Zod schema validation.
• sqlite-server/ — Test suite for a SQLite MCP server: 11 Playwright tests, 14 eval dataset cases.
• basic-playwright-usage/ — Minimal Playwright patterns.

Evaluation manifests

For projects with multiple datasets, use an evaluation manifest. Dataset paths are shorthand for tagged file sources, while hosts, metrics, judges, result stores, and other extensions resolve through public registries:

{
  "name": "tool-selection-search",
  "datasets": [{ "type": "file", "path": "evalsets/search.json" }],
  "servers": [
    {
      "transport": "http",
      "serverUrl": "https://example.com/mcp",
      "label": "prod"
    }
  ],
  "host": { "type": "sdk" },
  "metrics": ["passed"],
  "results": { "store": { "type": "file", "directory": ".mcp-test-results" } }
}

Run one manifest or a bounded batch:

npx mcp-server-tester run \
  --manifest eval-manifest.json \
  --plugins ./plugins \
  --dry-run

npx mcp-server-tester batch \
  --manifest-dir manifests \
  --workers 4 \
  --skip-existing \
  --dry-run

Use arms to compare server sets or host configurations. An arm can override servers, host options, tool maps, scenario templates, metrics, and judges. The canonical execution primitives remain EvalDataset, EvalCase, EvalMode, MCPConfig, and runEvalDataset.

Applications can register extensions with registerDatasetSource, registerHost, registerJudge, registerMetric, and registerResultStore. Secrets remain environment-variable or plugin-owned runtime inputs and do not belong in committed manifests.

Known Limitations

These MCP protocol features are not currently supported. These are deliberate scope decisions, not bugs:

• MCP resources (listResources, readResource)
• MCP prompts (listPrompts, getPrompt)
• Server-to-client notifications
• Streaming tool responses (callTool waits for the complete response)

If any of these affect your use case, please open an issue.

License

MIT