The agent restates the goal, decides the target host and asks for a site-specific recipe before it navigates anywhere, which is the difference between a run that survives a cookie banner and one that stalls on it. The other half of this server is the part most browser tools leave out: after a run fails you can read Browserless's own record of what was attempted, why it stopped, how long it took and what it cost, instead of guessing from a timeout.
A hosted client for Browserless: page export and scraping, site mapping and crawling, a persistent agent session with its own command loop, web search, Lighthouse audits, custom Puppeteer execution, saved authentication profiles, site-specific recipes, and the account's own usage and request logs. Nothing is installed — a client points at the hosted URL with a token.
- browserless_agent drives a persistent session under a snapshot-plan-act-resnapshot loop, addressing elements by refs taken from the snapshot, with multi-tab work, captcha solving, file upload and download, and an optional residential proxy fixed at session creation.
- browserless_skill loads an on-demand recipe for a page mechanic that usually derails a run — shadow DOM, cookie consent banners, modals, captchas, dynamic content, empty snapshots, screenshots, tabs — and recipes also inject themselves into the agent's replies when their triggers fire.
- Scraping at whichever scope you need: browserless_smartscraper for a single page as markdown or HTML, handling JavaScript-heavy pages and anti-bot measures; browserless_map to discover a site's URLs through sitemaps and link extraction; browserless_crawl to follow links to a set depth with path filters and per-page scrape options.
- browserless_export fetches a URL and returns its native format — HTML, PDF, image — with the content type detected for you.
- browserless_search searches web, news and images with geo-targeting and time filters, and can scrape each result.
- browserless_performance runs a Lighthouse audit returning accessibility, best practices, performance, PWA and SEO scores, filterable by category or checked against budgets.
- browserless_function runs custom Puppeteer JavaScript on the Browserless cloud against a page object, returning a data and type pair, so a real MIME type comes back as a proper content block rather than base64 text.
- browserless_profiles lists the saved authentication profiles — logged-in browser state as cookies and storage — with cookie and origin counts, so a task needing a signed-in browser can name one.
- browserless_account, browserless_usage and browserless_logs read the plan, unit balance and billing period, consumption broken down by successes, errors, timeouts, queueing, peak concurrency and proxy bytes, and Browserless's own record of what each recent request attempted, why it stopped and what it cost.
- browserless_sessions shows the browsers running now, persistent sessions on dedicated workers, and recorded replays — and can download a replay as a self-contained player page that renders with no network.
A Browserless API token, on a free tier or a paid one. Nothing is installed: point your client at the hosted server and pass the token in an Authorization header, or as a query parameter for clients that only accept a URL. A header also selects a regional endpoint. Some proxy options — state, city, named presets — are plan-gated, and a token without the plan is refused rather than silently ignored.
One command plus a key — npx -y @browserless.io/mcp, then supply credentials