Browser-Use Agent

Any AI agent that can control a Chromium browser can use espctl through esphome.cloud/mcp/esp-idf — without installing anything. No binary, no package, no PATH.

This page covers setup for agents like browser-use, computer-use, or any framework that drives a browser via CDP, Playwright, or Puppeteer.

Requirements

Configuration

No MCP server config needed. The agent opens a browser tab instead of running a binary. Point your agent at:

https://esphome.cloud/mcp/esp-idf

If your agent framework has a “start URL” or “initial page” setting, use that URL. If it needs a task description, tell it:

Open https://esphome.cloud/mcp/esp-idf in Chrome. Sign in if prompted. Click Connect. Then follow the build instructions.

Build flow

The agent follows this sequence in the browser:

1. Navigate to esphome.cloud/mcp/esp-idf.
2. Sign in if a sign-in prompt appears.
3. Click Connect — wait for the green dot.
4. Pick target chip from the dropdown (esp32, esp32s3, …).
5. Pick IDF version (optional — the default works).
6. Pick build type (release or debug).
7. Click Build — logs scroll live in the panel below.
8. Wait for the build to finish (status changes to succeeded or failed).
9. Read results — click Size Report, SBOM, or Diagnostics for post-build analysis.
10. Download firmware — click the download icon on the firmware card.

Flash flow (optional)

If the agent has access to a USB-connected ESP device:

1. Switch to the Flash tab.
2. Click Connect — pick the USB device from the port list.
3. Click Flash.

Note: Web Serial requires the browser to have USB access. If the agent runs in a headless or sandboxed environment, flashing may not be available.

Monitor flow (optional)

The Monitor tab works without signing in or connecting to the build server:

1. Switch to the Monitor tab.
2. Click Open Monitor — pick the USB device.
3. Pick a baud rate (115200 default).
4. Read serial output.

Check it’s working

After the agent navigates to the page and clicks Connect, it should see:

• A green connection indicator
• A Tools Inspector panel listing available tools
• The Build configuration controls (target, version, build type)

If the agent sees a sign-in prompt instead, it needs to complete sign-in first.

Compared to local MCP

Use browser MCP when your agent can’t install binaries. Use local MCP when your agent has shell access and you want the tighter MCP protocol integration.

See also

• MCP Console — full reference for the browser MCP page.
• Claude Code — local MCP setup via espctl mcp serve.
• Tool Reference — all 40 tools.