Browsir AI
https://github.com/mksglu/browsirai- Category
- AI Agents
- Rank
- No. 1566Tools index
Previous survey · No. 1573 ·
- Pricing
- Open Source
- Type
- AGENT
- Builder
- mksglu
- GitHub
- 45 stars
- Latest release
- v0.2.1
- Date
About
Your browser, your sessions, your agent — CDP-based browsing agent that respects your real sessions.
What it does
Browsir AI gives agents and terminal users structured control over Chrome. It exposes navigation, form interaction, accessibility snapshots, screenshots, console output, network capture, request interception, saved browser state, and component source inspection through MCP and command-line interfaces.
Why it's ranked here
The tool combines practical browser automation with unusually strong debugging coverage. Accessibility references reduce dependence on coordinates, while console capture, network interception, pixel comparison, and framework-aware source inspection support real development work. Its session handling and automatic upgrade behavior deserve careful evaluation before adoption.
What's good
Cookie transfer happens locally between SQLite databases, keeping cookie values outside MCP messages according to the documentation. Sensitive headers, common secret fields, bearer tokens, and JWT-shaped strings receive additional redaction. Bounded event storage preserves recent console and network activity without unlimited memory growth.
Tradeoffs
It depends on Chrome or a compatible Chromium browser, Node.js, and remote debugging access. Starting a separate browser can involve a temporary profile and copied cookies. Automatic update checks contact the npm registry, and upgrades may clear the npm cache or modify a global installation.
How to use it well
Use it for agent-assisted web debugging, authenticated browser tasks, interface testing, and terminal-driven investigation where existing sessions matter. Prefer text snapshots for routine interaction and visual captures for layout questions. Do not treat browser automation as a replacement for stable service APIs or dedicated credential management.
Technical notes+
package.json defines an ESM Node.js package requiring Node 18 or newer, with npm executable and module entry surfaces. src/bin.ts passes command arguments into the dispatcher. src/cli.ts starts a stdio MCP server by default, routes install and doctor commands, dynamically loads standalone CLI commands, and globally suppresses unhandled rejection and uncaught exception events. src/server.ts registers the browser tools with the MCP SDK. src/config.ts validates layered defaults, JSON configuration, and environment overrides with Zod. src/ref-system.ts rebuilds sequential accessibility references and maps them to CDP backend node identifiers. src/event-buffer.ts implements bounded ring storage with eviction statistics. src/redactor.ts masks sensitive headers, selected JSON keys, bearer tokens, and JWT-shaped text. src/upgrade.ts checks npm with a one-hour cache and can launch background npm maintenance. src/chrome-launcher.ts contains macOS, Linux, and Windows browser discovery plus remote-debugging launch logic.
Observed
- License
- AGPL-3.0
- Primary language
- TypeScript targeting an ESM Node.js package
- Install surface
- Published npm package designed for npx use, with no global installation required for quick start
- Interfaces
- MCP server over stdio and standalone command-line interface
- Runtime requirement
- Node.js 18 or newer
- Browser support
- Discovery logic covers Chrome, Chromium, Microsoft Edge, and Brave candidates across macOS, Linux, and Windows
- Configuration
- Validated JSON configuration with environment-variable overrides
- Package contents
- Published package includes built output and skills directories
Read from README.md, package.json, src/cli.ts, src/server.ts, src/bin.ts, src/config.ts, src/doctor.ts, src/errors.ts, src/install.ts, src/upgrade.ts, src/version.ts, src/redactor.ts, src/ref-system.ts, src/event-buffer.ts, src/chrome-launcher.ts.
What it can do
Control browser sessions programmatically
Browser session and automation commands → Executed browser actions and interactions
Access existing browser sessions
Active browser session data → Connected agent with session context
Automate web interactions using CDP protocol
Web pages and interaction instructions → Automated clicks, form fills, and navigation
Extract data from web pages
Web page content and extraction parameters → Structured data or specific page elements
Monitor browser activity
Browser session events and page changes → Activity logs and session data
Execute custom browsing workflows
Workflow definitions and target websites → Completed automated browsing tasks
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.