Cool Test MCP

An automated testing MCP (Model Context Protocol) server. It is lightweight — no extra runtime dependencies, single entry point, instant to start — and plugs into any MCP-capable agent (Claude Desktop, Cursor, opencode, etc.) with a few lines of config.

Features

• Lightweight & easy to adapt — no extra runtime deps, runnable via npx, one-line config to connect to your agent tools

• Conversion — turn test cases in any format into a fixed JSON template (.cooltest/)

• Case-by-case testing — read / test / write cases one by one through MCP tools, avoiding direct JSON file I/O that wastes tokens

• Review flow — cases that cannot be tested or judged are automatically set to review, left for human review

• Visual report — a local web page shows all case results and supports editing status and notes

Related MCP server: mcp-playwright-test

Requirements

• Node.js 18 or newer

• Claude Desktop, Cursor, opencode, or any other MCP client

Getting started

No local install needed — the server runs directly via npx. Register it in your agent's MCP configuration.

Standard config works in most of the tools:

{
  "mcpServers": {
    "cool-test": {
      "command": "npx",
      "args": [
        "-y",
        "cool-test-mcp@latest"
      ],
      "cwd": "/absolute/path/to/your/project"
    }
  }
}

cwd points to your project root; .cooltest/ will be created there. For clients without a per-server cwd, the folder is created in the MCP process's working directory.

Use the Codex CLI to add the server:

codex mcp add cool-test npx "-y cool-test-mcp@latest"

Alternatively, create or edit ~/.codex/config.toml:

[mcp_servers.cool-test]
command = "npx"
args = ["-y", "cool-test-mcp@latest"]

For more information, see the Codex MCP documentation.

Follow the MCP install guide, and add to claude_desktop_config.json:

{
  "mcpServers": {
    "cool-test": {
      "command": "npx",
      "args": [
        "-y",
        "cool-test-mcp@latest"
      ],
      "cwd": "/absolute/path/to/your/project"
    }
  }
}

Go to Cursor Settings -> MCP -> Add new MCP Server. Name it to your liking, use command type with the command npx -y cool-test-mcp@latest. Alternatively, add to .cursor/mcp.json:

{
  "mcpServers": {
    "cool-test": {
      "command": "npx",
      "args": [
        "-y",
        "cool-test-mcp@latest"
      ]
    }
  }
}

Create a .trae/mcp.json file in your project root (Cursor-compatible mcpServers schema):

{
  "mcpServers": {
    "cool-test": {
      "command": "npx",
      "args": ["-y", "cool-test-mcp@latest"]
    }
  }
}

For more information, see the Trae MCP documentation.

Zed uses the context_servers key (not mcpServers). Add to ~/.config/zed/settings.json:

{
  "context_servers": {
    "cool-test": {
      "command": "npx",
      "args": ["-y", "cool-test-mcp@latest"]
    }
  }
}

For more information, see the Zed MCP documentation.

Follow the MCP Servers documentation. For example in ~/.config/opencode/opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "cool-test": {
      "type": "local",
      "command": [
        "npx",
        "-y",
        "cool-test-mcp@latest"
      ],
      "enabled": true
    }
  }
}

The MCP configs for each agent tool above have not all been individually verified. If you find any issues or missing configs, we welcome PRs to add or fix them.

Usage

Trigger the full test flow

Use Cool Test for <test case address>

The LLM will: check whether your agent has browser automation capability (Playwright MCP etc.) → convert the cases → test case by case → open the report.

View the report

Use Cool Test to view <address>

MCP Tools

Contributing

Contributions are welcome. To set up a local environment:

# Clone
git clone https://github.com/CoolTea001/cool-test-mcp.git
cd cool-test-mcp

# Install
npm install

# Build (compile TS + copy the report script to dist)
npm run build

# Test (end-to-end check, full tool flow via MCP client)
node test-e2e.mjs <temp dir>

License

MIT