DEV Community

power zhong
power zhong

Posted on

Tried `ChromeDevTools/chrome-devtools-mcp`: A Practical Browser Debugger for Coding Agents

#nextjs #showdev #sideprojects #webdev

Tried ChromeDevTools/chrome-devtools-mcp: A Practical Browser Debugger for Coding Agents

I test a lot of “AI developer tools,” but ChromeDevTools/chrome-devtools-mcp immediately felt practical. It exposes Chrome DevTools capabilities through MCP, allowing coding agents to inspect pages, read console errors, analyze network requests, measure performance, and interact with a real browser.

That explains the current traction: +215 GitHub stars today. Instead of asking an agent to guess why a frontend is broken, the agent can check the actual runtime state. For solo builders, this is especially useful when shipping React, Next.js, or Dockerized dashboards without a dedicated QA person.

A minimal MCP configuration looks like this:

{
  "mcpServers": {
    "chrome-devtools": {
      "command": "npx",
      "args": [
        "-y",
        "chrome-devtools-mcp@latest"
      ]
    }
  }
}
Enter fullscreen mode Exit fullscreen mode

My preferred architecture is:

Cursor / Claude Code
        │ MCP
        ├── chrome-devtools-mcp
        │       └── Chrome with remote debugging
        │
        └── AI API gateway
                └── claude-fable-5
Enter fullscreen mode Exit fullscreen mode

For the model side, I can point an OpenAI- or Anthropic-compatible client at a custom gateway rather than hard-coding a vendor endpoint:

export ANTHROPIC_BASE_URL="https://b-lost.com/v1"
export ANTHROPIC_API_KEY="your-b-lost-key"
export ANTHROPIC_MODEL="claude-fable-5"
Enter fullscreen mode Exit fullscreen mode

The B-Lost Universal Relay advertises 20% off official list pricing, plus a 100% first-deposit bonus from $10 minimum. More importantly for repeated browser-debugging prompts, its native Anthropic /v1/messages support includes full Prompt Caching, with 90% discounts on cache hits. That can matter when the agent repeatedly receives the same project rules, tool instructions, and frontend context.

This is not a replacement for testing discipline, but it makes browser debugging observable and automatable. For an indie hacker trying to ship faster while keeping monthly tooling costs low, that is a compelling combination.

Top comments (0)