# devharness MCP server

> Connects AI assistants to Chrome DevTools Protocol to debug, inspect variables, monitor network traffic, and automate browser interactions.

- Repository: https://github.com/InDate/devharness
- Page: https://hostedmcp.org/servers/indate/devharness
- Transport: stdio
- Stars: 19
- License: MIT

## Install

### Claude Code

```sh
claude mcp add devharness -- npx -y devharness
```

### Cursor (`~/.cursor/mcp.json`)

```json
{
  "mcpServers": {
    "devharness": {
      "command": "npx",
      "args": [
        "-y",
        "devharness"
      ]
    }
  }
}
```

### VS Code (`.vscode/mcp.json`)

```json
{
  "servers": {
    "devharness": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "devharness"
      ]
    }
  }
}
```

### VS Code CLI

```sh
code --add-mcp '{"name":"devharness","type":"stdio","command":"npx","args":["-y","devharness"]}'
```

### Claude Desktop (`claude_desktop_config.json`)

```json
{
  "mcpServers": {
    "devharness": {
      "command": "npx",
      "args": [
        "-y",
        "devharness"
      ]
    }
  }
}
```

### Codex

```sh
codex mcp add devharness -- npx -y devharness
```

### Codex config.toml (`~/.codex/config.toml`)

```toml
[mcp_servers.devharness]
command = "npx"
args = ["-y", "devharness"]
```

### Gemini CLI

```sh
gemini mcp add -s user devharness npx -- -y devharness
```

### Gemini settings.json (`~/.gemini/settings.json`)

```json
{
  "mcpServers": {
    "devharness": {
      "command": "npx",
      "args": [
        "-y",
        "devharness"
      ]
    }
  }
}
```

## FAQ

### What does devharness allow an AI agent to do?

It gives the agent its own Chrome window to navigate apps, fill forms, click buttons, and read errors from the browser console and server logs.

### Can the agent debug Node.js code?

Yes, it can set breakpoints, step through code, and inspect the call stack and variables in both Chrome and Node.js processes started with --inspect.

### How can I use the recorded steps from an agent?

The steps taken by the agent can be saved as a sequence to be replayed later or exported as a Playwright or Puppeteer test.

### What are the system requirements for setup?

You must have Google Chrome installed and Node.js version 18 or later.