Haden Browser

AI-first browser exposed through a local MCP server that returns compact goal-based page views instead of full screenshots or raw page source on every turn. It uses Playwright and lazy image retrieval to lower browser-agent context costs while still supporting semantic action references, article reading, form interactions and computed-style inspection. Public GitHub repository created August 19, 2026.

MCP unverified

Integration

Transport	`stdio`
Auth	`none`
Endpoint	https://github.com/Haden-Min/haden-browser
Install	`npx -y haden-browser`

Use Cases

01	Give an AI agent browser-control tools that return compact semantic views rather than full page dumps by default
02	Read articles, submit forms and inspect computed styles through goal-specific MCP tools
03	Fetch image pixels only when explicitly needed while preserving alt-text and action references in normal page context