Almost done testing it and it will soon be in bumped version on https://t.co/gIcsDvupY1.
Amongst other things that i'm trying is a Raycast extension - or rather a way to do "pagerunning" via spotlight-kind of interface. Not entirely sure if that is THAT useful yet.
Doing a bigger pass on making Pagerunner working gracefully with profile locks, crashes, and even sleep.
There is a fairly smart system which checkpoints sessions, detects failure modes and restore when everything goes wrong. It is quite handy.
Daemon + browser sessions are really good now together with Mac OS Bar as well.
I've been building Pagerunner โ browser MCP using Chrome DevTools Protocol. Named encrypted profiles. Local PII stripping. Daemon mode. 27 tools. Rust. Open source. Full analysis + benchmarks: https://t.co/K2hmommASS https://t.co/M2VngqgygH
Anthropic just shipped browser automation in Claude Code. 14M views on the announcement. Here's what the benchmarks actually say about screenshot-based automation ๐งต
DOM-based tools won 4 of 5 browser tasks in head-to-head benchmarks.
Computer Use: "click the blue button, third from left"
CDP: click("button.submit").
One is guessing. One is reading.
It's OCR-ing a spreadsheet instead of opening the CSV.
Just launched pagerunner-skill โ real Chrome automation for AI agents, using your existing login sessions. No API tokens, no re-auth.
What it does:
โ Controls your real Chrome with existing cookies & sessions
โ PII anonymization โ emails/phones get tokenized before reaching the LLM โ Multi-agent coordination via persistent daemon + KV store โ Audit logging for compliance โ Works with 27 browser automation tools
Get it in one command: npx skills add Enreign/pagerunner-skill
Works in Claude Code, Cursor, Windsurf, Roo, Goose, Continue, Kilocode + 20 more tools out of the box.
4 use cases: ๐ Solo dev โ edit code โ screenshot โ verify in real browser
๐ฑ Power user โ trigger browser tasks from your phone via agents ๐ Security โ PII never reaches the LLM, local NER, audit trail โ๏ธ Server โ headless cron automation with agent-to-agent handoff
https://t.co/oK4zcWHCIX
Pagerunner v0.1 live ๐
Agent-first browser orchestration MCP server for Claude. Drive real Chrome via CDP with sessions, snapshots, KV store, PII anonymization & audit logging.
https://t.co/gIcsDvupY1
Agents crush small tasks. on complex ones? ~25% autonomous completion.
most teams don't account for this when scoping AI-assisted work.
just shipped v0.5.0 of progressive-estimation โ because "it depends" is not an estimate.
https://t.co/PAh7lJzOFT
Estimation formulas built for humans don't work when half your sprint is AI agents.
Built a fix: https://t.co/PAh7lJzOFT
Early and rough. Bug reports welcome.
I'm building my dream video editor with Claude AI
11k lines of code, 685 commits, 3 full weeks. Cursor, Tailwind, React, Firebase. Web-based, cloud saves, versions, can edit on mobile. This video was recorded using the tool.
Never in a million years I thought I was capable but here we are. AI is a multiplier.