stealthy-auto-browse

One container runs Camoufox, a graphical display, an HTTP API, an MCP server, and a noVNC viewer. A client sends browser actions. The browser keeps its tabs and session state, and you can watch or take over through noVNC.

Use it for authorized browser automation, compatibility testing, and QA. It offers both Playwright element actions and real X11 pointer, keyboard, and wheel input. Neither mode guarantees that a site will accept the session or classify it as human.

Start here

docker run -d --name browser \
  -p 127.0.0.1:8080:8080 \
  -p 127.0.0.1:5900:5900 \
  psyb0t/stealthy-auto-browse:latest

Open http://localhost:5900/ for the viewer. Send HTTP commands to http://localhost:8080/, or connect an MCP client to http://localhost:8080/mcp/. This example is loopback-only and unauthenticated. See installation and security before making it reachable from another machine.

curl --fail-with-body http://localhost:8080/ \
  -H 'Content-Type: application/json' \
  -d '{"action":"goto","url":"https://example.com"}'

What it supports

Area	Features and reference
Browser control	Navigation, tabs and windows, dialogs, permissions, geolocation, cookies, and storage.
Element targeting	CSS, XPath, accessible roles, iframe chains, open shadow roots, and reusable element references.
Input	DOM actions, OS-level mouse gestures, typing, scrolling, and visible reading pauses.
Capture	Text, HTML, accessibility snapshots, PNG screenshots, MP4 recording, network logs, and HAR export.
Browser events	Bounded event history, waits, SSE delivery, and human handoff.
Clipboard and media	Text, HTML, PNG, file lists, file-backed camera and microphone streams, and runtime source switching.
Challenges	Best-effort detection, optional solver providers, and human takeover.
Workflows	Atomic multi-step requests and bounded YAML control flow, plus URL-triggered page loaders.
Session identity	Persistent profiles and fingerprint upgrades, input models and their limits.
Clients	HTTP and MCP setup, dedicated MCP tools, and published agent integrations.
Replicas	HAProxy queues, sticky sessions, and Redis cookie sharing. Browser state other than cookies remains per replica.

Operate and contribute

Configuration owns settings and defaults. Troubleshooting covers startup, authentication, browser recovery, and cluster problems. Development explains the source layout, build, deterministic tests, and opt-in external checks.

The source repository contains the license and changelog.