Here's the fastest way to set up Lightpanda
Point your Claude, Cursor, or any MCP client config at the binary
Your agent gets markdown, semantic tree, interactive elements + structured data
https://t.co/oU0cO1OQWP
@reformedot@MarcioK@gregpr07 Stealth is harder when starting from scratch, no argument there.
BUT engine choice doesn't have to be binary. Why not offer Lightpanda as an alternative like agent browser and Hermes?
https://t.co/LUb6rbrzyF
We benchmarked Lightpanda for agents and learned the tool surface you hand the model matters as much as the engine underneath it.
Here's what we're shipping because of it 🧵
https://t.co/PpSQ6TteSp
Lightpanda now implements the CookieStore web API 🍪
It's the modern async way to read and write cookies, instead of parsing document.cookie strings by hand.
We also cleaned up cookie ownership under the hood, which removes a double-free risk on certain error paths so sessions stay stable when things go wrong.
Slack, crypto exchanges, live dashboards: all push real-time data through a WebSocket inside a Web Worker.
Lightpanda couldn't run that path until today.
Below: a Worker hitting Binance, BTC trades streaming into the terminal.
PR: https://t.co/Ow5aCaU9Ek
Code: https://t.co/rUQG9a4s0C
i just asked my Hermes Agent what the 5 most used skills are in our multi-agent system so far.
anw this entire Hermes setup is mostly just for my side projects and hobby stuff 😂
1️⃣ cloudflare-mcp
DNS, Workers, R2, CDN. Almost every deploy flows through this.
2️⃣ porkbun-api
31 MCP tools for domain registration, DNS records, SSL.
3️⃣ prediction-market-trading
Polymarket + Kalshi via @Predexon. Trade execution, pair detection, smart wallet profiling.
4️⃣ getxapi-twitter-data-api
X/Twitter extraction. Search, timeline, profiles. ~$0.001/call with basically no rate limits.
5️⃣ lightpanda-batch-scraping
Headless browser stack for markdown extraction, JS-heavy scraping, parallel crawling, fallback strategies.
Chrome shipped WebMCP at I/O. A new browser API where sites declare tools for agents to call directly.
Catch from Google’s docs: it doesn’t work headless.
We fixed that in Lightpanda
https://t.co/7InJ7aeIj9
The WebMCP API will enter origin trial starting in Chrome 149 → https://t.co/uQdxi2C6FP
It's a proposed web standard that lets you expose structured tools, like JS functions and HTML forms, to browser-based agents, enabling them to know exactly how and where to interact with your site.
#GoogleIO
4/ Lightpanda fetch gets a --json flag. HTTP status and headers come back in JSON so the one-shot fetch fits into scraping pipelines where you need to know if you got a 503.
New in Lightpanda this week 🧵
1. Accessibility.queryAXTree over CDP
2. External stylesheets fetched and parsed
3. fetch gets a --json flag
All in the latest nightly: https://t.co/DW2VL6K8YG
3/ external stylesheets are now fetched and parsed behind --enable-external-stylesheets. Visibility and computed style checks now reflect what's actually rendered on the page.