Give your AI agent a real browser.

Five lines to a working agent

Install, one MCP block, one REST call. Everything else on this page is detail.

# 1. Install and launch (Node 18+), or use the native installer from /download
npx lumabrowser start

# 2. MCP: Settings > Connect your agent (Claude Code, Codex, OpenCode, Cline),
#    or Settings > Export MCP Configuration for any other host (needs Node.js)

# 3. REST: open a page, then click with an LLM fallback if the selector is stale
curl -X POST localhost:3000/api/browser/tabs -H 'Content-Type: application/json' -d '{"url":"https://news.ycombinator.com"}'
curl -X POST localhost:3000/api/browser/tabs/0/click -H 'Content-Type: application/json' -d '{"selector":".titleline > a","llmFallback":"Click the first story link"}'

Ports: REST 3000, plus WebDriver 9515 and CDP 9222, which are off by default and turned on in Settings. Headless in Docker: clone the repo and run LUMA_VNC_PASSWORD=change-me docker compose up -d (see Docker). The step-by-step version with config file locations is the agent quickstart.

One click, or one block of JSON

LumaBrowser ships an MCP server. In Settings, “Connect your agent” writes the entry into Claude Code, Codex, OpenCode and Cline for you. For any other host (Claude Desktop, Cursor, OpenClaw), “Export MCP Configuration” saves luma-mcp-config.json with the block below.

{
  "mcpServers": {
    "luma-browser": {
      "command": "node",
      "args": ["<install dir>/resources/app.asar.unpacked/mcp-server.js"],
      "env": { "LUMA_API_PORT": "3000" }
    }
  }
}

The selector breaks. Your agent doesn't.

Modern frontends ship CSS-in-JS class names like css-8xk2m9 that change on every build. Most MCP browser tools delegate to Puppeteer or Playwright and crash with element not found. LumaBrowser's selector tools take an optional llmFallback; if the CSS misses, an LLM resolves the description against the live DOM and the call succeeds.

click({ selector: ".product-buy-btn" })
  → Error: element not found

click({
  selector: ".product-buy-btn",
  llmFallback: "the primary buy button
                under the price"
})  → ok

Stop dumping raw HTML into your context

If your agent is reading entire DOM trees to figure out where to click, you're burning tokens on every turn. LumaBrowser tries the cheap paths first: an accessibility match that costs no tokens, then a local cache of selectors the model already resolved. The model is only asked when both miss, and it sees a compact list of up to 80 interactable elements, not the raw HTML.

What your agent can do

A focused set of MCP tools, not a 60-tool kitchen sink. Every interaction tool that targets the DOM accepts an llmFallback description.

Where LumaBrowser fits in the stack

There are a lot of ways to give an agent a browser right now. Here's how the local-desktop-app approach compares to the alternatives engineers usually try first.

Pricing as listed by each vendor at time of publication. Cloud headless-browser pricing is metered, Browserbase by browser-hours, Browserless by 30-second “Units”, so cost scales with how often your agent loops. Higher tiers (Browserbase Startup at $99/mo, Browserless Starter at $140/mo annually) raise the included quota and concurrency.

This comparison reflects publicly available pricing and feature information gathered to the best of our knowledge from each vendor's public materials. Vendors update plans frequently and we're a small team; if anything here looks wrong, please email [email protected] with the correction and a source, and we'll update the page.

Compatibility & BYOK

Bring your own MCP host. Bring your own model.