JohnXu22786/browser-automation
Real-browser automation for dsh via an MCP server (Playwright core): navigate, click, fill, type, screenshot, and act on accessibility snapshots with numbered references - all 22 web_* tools re-exposed to the harness over stdio.
Web Bridge — a browser-automation MCP server that gives agents a real browser's eyes and hands. Its core is the accessibility-tree snapshot (web_snapshot): the page is rendered into a structured text tree with numbered interactive elements (refs), so the agent can click, fill, hover, scroll and switch tabs precisely without a vision model or selector guessing. 22 web_* tools cover multi-tab control, screenshots (viewport/full-page/element) and sanitized in-page JS execution.
Install
dsh plugin --profile demo add github:JohnXu22786/browser-automationREADME-documented install into the demo profile from GitHub (replace demo with your profile). The repo ships a Cordis bundle: package.json declares dsh.bundle.patch -> cordis.patch.yml and index.js bridges to the MCP server. In a source checkout dist/ is gitignored so the bridge runs npm run build once on first load; installed npm packages ship dist/ and need no build. Browsers require npx playwright install chromium on first use.
Compatibility
TypeScript (Node.js >= 20) speaking MCP over stdio; browser engines Chromium / Firefox / WebKit (Chromium default). First use requires npx playwright install chromium.
Details
- Repo: JohnXu22786/browser-automation
- Category: Tools & Capabilities
- Stars: 1
- Version: GitHub main (README-documented); npm package not referenced by the README
- Last push: 2026-08-17
- First seen: 2026-08-16
Recent updates
README documents the Cordis bundle bridge as an alternative to the dsh-plugin.json MCP-server manifest; profile-level settings map to the same WEB_BRIDGE_* environment variables.
FAQ
- How do I install browser-automation?
- Run: dsh plugin --profile demo add github:JohnXu22786/browser-automation (use your own profile name), then install a browser engine with npx playwright install chromium.
- Does it need a vision model?
- No. It works from the accessibility-tree snapshot, which renders the page as a structured text tree with numbered interactive elements, so plain text models can drive it precisely.
- Which browsers are supported?
- Chromium (default), Firefox and WebKit — install the engine with the matching npx playwright install <engine> command.
Alternatives
zhaoolee/notes · Vladimir-Human/ru-marketplace-mcp#dsh · DDDMUC/dsh-free-search