Credentials. Agent mode reads model API keys from your environment; scope them and watch usage.
cost. LLM calls in agent mode cost tokens; replaying saved scripts does not.
Who it is forDevelopers building agents that need to read and click through websites.
The problemAgent browsing with full Chrome is slow and costs tokens on every replay.
The outcomeAn agent tool that browses pages, plus a token free PandaScript you can rerun.
What you need first
- Linux or macOS (x86_64 or aarch64); Windows only through WSL2; Linux binaries need glibc
- Install with `brew install lightpanda-io/browser/lightpanda`, a nightly binary, Nix or `docker run -d --name lightpanda -p 127.0.0.1:9222:9222 lightpanda/browser:latest`
- puppeteer-core or Playwright if you connect your own scripts over CDP
Steps
- Add an mcpServers entry named lightpanda with command set to the lightpanda path and args ["mcp"]
- Or try agent mode with
./lightpanda agent --task "top story on news.ycombinator.com?", which auto detects an API key from the environment - Run
/savein a session to export a PandaScript, then replay it with./lightpanda run session.js - For several agents, start
lightpanda mcp --port 9223so each connection gets its own browsing session
Prompt for your agent
Using lightpanda-io/browser, add the lightpanda MCP server to my agent config, then use it to open <a page> and extract its title and links.
Caveat. Use `--no-llm` for a plain REPL without any model.
Use this in your AI agent
Ask your agent
Using the Deduped MCP, get the setup steps for lightpanda-io/browser and "Give an AI agent a Browser through MCP or agent mode"
Add Deduped to Claude Code once
claude mcp add --transport http deduped https://dedupedai.com/mcp
Every Monday: the 5 hottest repos and exactly how to use them. Free, one email a week, unsubscribe anytime.
Get them freeShare: X LinkedIn Hacker News Email Copy link
Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer