Browsegrab
Equip local or cost-sensitive agents with Playwright browsing that uses accessibility trees and MarkGrab for lower token use.
Overview
io.github.ArkNill/browsegrab is a Build-phase MCP server that provides token-efficient Playwright browser automation via accessibility trees and MarkGrab.
What is this MCP server?
- Playwright-driven browser agent over stdio MCP (PyPI browsegrab v0.1.1)
- Accessibility-tree navigation to cut token burn vs raw DOM dumps
- MarkGrab capture integration for structured page state
- Positioned for local LLMs and token-efficient agent loops
- GitHub source: QuartzUnit/browsegrab
- Server version 0.1.1
- PyPI identifier browsegrab with stdio transport
- Documented stack: Playwright + accessibility tree + MarkGrab
What problem does it solve?
Browser MCP tools often flood context with huge DOM snapshots, making local LLMs slow, expensive, and unreliable.
Who is it for?
Indie builders automating QA or research flows with Claude Code, Cursor, or local models under tight token budgets.
Skip if: Teams that only need static HTTP fetch, lack Playwright install appetite, or want fully managed cloud browser farms.
What do I get? / Deliverables
After install, your agent can navigate and inspect pages through lean accessibility-tree and MarkGrab outputs suited to automation loops.
- Interactive browser sessions driven from the agent
- Lean page representations via accessibility tree and MarkGrab
- Reusable automation for smoke tests and research crawls
Recommended MCP Servers
Journey fit
Build → agent-tooling is the first journey stop where you add browser capabilities to the coding agent itself. Agent-tooling reflects Browsegrab’s core: token-efficient browser control as MCP tools, not a finished end-user SaaS feature.
How it compares
Token-efficient Playwright MCP agent, not a static scraper skill or hosting marketplace.
Common Questions / FAQ
Who is io.github.ArkNill/browsegrab for?
Solo developers who want coding agents to control a real browser with Playwright while keeping prompts smaller via accessibility trees and MarkGrab.
When should I use io.github.ArkNill/browsegrab?
Use it during Build agent-tooling, Validate prototyping, or Ship testing whenever you need interactive browser steps instead of one-shot HTTP reads.
How do I add io.github.ArkNill/browsegrab to my agent?
Install the PyPI package browsegrab 0.1.1, ensure Playwright browsers are available, register stdio MCP in your host, then invoke Browsegrab tools from your agent config.