Browserless Review

One unit equals 30 seconds of browser time. Free gives you 1,000 units and two concurrent browsers, no card required. Starter runs $140/month annually for 180,000 units and 40 concurrent browsers — that math is easy to pencil against actual job volume.

The self-hosted Docker image under SSPL-1.0 is the real differentiator against Browserbase and Bright Data, both named on their own comparison pages. Stealth, CAPTCHA solving and proxies on every plan, including Free, is generous. SOC 2 Type II and a HIPAA BAA cover the compliance bar for regulated teams.

The tradeoff: six endpoints including /unblock and /smart-scrape, plus managed proxies, are cloud-only — self-hosting means losing the convenience layer. The /agent/run endpoint is still beta. Fine for engineers building scrapers today; less proven for teams betting heavily on autonomous agent workflows.

Unit-based pricing — 30 seconds of browser time per unit, 1,000 free monthly — is a cleaner cost model than per-seat or flat API-call pricing for anything running headless Chrome at scale. It maps cost directly to the thing that actually burns infrastructure: browser-seconds. That's a sign whoever built the billing layer understands what breaks budgets in this category.

The BrowserQL protocol and BAP SDK are the bigger architectural tell. Building a GraphQL automation layer on top of Puppeteer/Playwright compatibility means they're not just reselling Chrome instances — they're building a durable abstraction layer. If that protocol matures, you get structured extraction and conditional logic without hand-rolled selector logic; if it stalls, you've got a second DSL to maintain alongside your existing scripts.

The real long-term constraint: six REST endpoints including /unblock and /smart-scrape, plus managed proxies, run cloud-only. Self-hosting gets you Puppeteer, Playwright and core REST under SSPL-1.0, but the anti-bot tooling that's actually hard to replicate stays behind the managed product. Plan your architecture assuming that split persists for three years, not two.

Five tiers, four with real numbers attached. Free: 1,000 units, no card. Prototyping $25/month annual, 20,000 units. Starter $140, 180,000 units. Scale $350, 500,000 units. That's a clean unit-cost curve — $0.0007/unit at Scale, cheaper than $0.00125 at Prototyping. Volume actually buys something here.

Team of 50 running scraping jobs at moderate volume lands on Starter or Scale fast — 500,000 units is 166 hours of browser time a month, not a lot once CAPTCHA solves (10 units each) and proxy MB (6 units residential) start eating the pool. Year 3 cost depends entirely on how much proxy and CAPTCHA volume you run, not seat count. No seats here at all, which is its own kind of honest.

Enterprise is the one quote-only tier — GPU infrastructure, dedicated TAM, commercial self-host license. Fine, that's custom infrastructure pricing, not bait. Monthly billing, no term-length info I could find, so I can't score auto-renewal risk either way.

The connect()/connectOverCDP() swap is real — your existing test suite or scraper doesn't need a rewrite, just a new target. That's rare in this category where most 'drop-in' claims mean partial compatibility.

Where it gets interesting is BrowserQL and the MCP server. BrowserQL has its own IDE with autocomplete and query history, which signals a team that writes real automation, not just a REST wrapper over Chrome. The MCP server wiring into Claude, Cursor, VS Code and Windsurf is forward-looking for agent work, but /agent/run is explicitly beta — I wouldn't put it in a production pipeline yet, and the docs should say what breaks when it changes.

Session max time caps out at 15 minutes on Prototyping and 60 on Scale, which matters for anything doing long-running crawls — you'll hit reconnect logic sooner than you'd like. Self-hosted Docker keeps Puppeteer, Playwright and core REST working, but losing /unblock and /smart-scrape outside the cloud product means your local dev environment and your production environment behave differently. That's a day-3 fight waiting to happen.

This isn't a product you poke around in for fun, it's a thing you wire into a script and forget about until something breaks. That's the right posture for this category, and the dashboard reflects it: request logs filterable by endpoint, outcome, target URL and API key, plus session replay and a live debugger. That's the kind of screen you build when you've personally been paged at 2am wondering why a scrape failed.

First ten minutes is genuinely low-friction if you already write Puppeteer or Playwright, since it's a one-line swap to connect(). If you don't already know those libraries, you're doing homework, not onboarding, there's no beginner on-ramp here.

The free tier giving you 1,000 units with no credit card is a real try-before-you-buy, not a teaser. Month three is where BrowserQL and the typed BAP SDK start paying off, since raw scripts get brittle at scale. Mobile doesn't apply to this category, so I won't dock it for that. Self-hosted Docker is a genuine escape hatch if you ever distrust the cloud bill.

Six named competitors on the comparison pages. That's not a red flag by itself — it's a crowded lane, and crowded lanes have both long survivors and quiet shutdowns. What separates them is usually execution, and the unit-cost dashboard plus session replay suggests Browserless is paying attention to the stuff that normally gets neglected until a billing complaint forces it.

Self-hosted Docker under SSPL-1.0 is a real exit path — better than most managed scraping shops offer. But six REST endpoints and the managed proxies stay cloud-only, so the self-hosted version isn't actually the same product, it's a reduced one. That's the kind of asterisk marketing copy doesn't lead with.

SOC 2 Type II and a HIPAA BAA are concrete, verifiable claims, not aspirational ones. /agent/run being in beta is honestly labeled, which counts for something. Whether BrowserQL and the MCP server are differentiated enough to beat Bright Data on its home turf — unclear from here.