Browserless
Also known as: Browserless.io, BrowserQL
Headless browser infrastructure for automation and AI agents: managed Chrome, Firefox and WebKit over WebSocket, BrowserQL, REST APIs and an MCP server with a 136-skill catalog, with self-hosting into a VPC or air-gapped network.
Browserless is headless browser infrastructure for automation and AI agents. It runs managed Chrome, Firefox and WebKit that existing Puppeteer, Playwright or Selenium code connects to over WebSocket, alongside BrowserQL, a GraphQL-based language for multi-step automation with stealth and structured extraction, and REST APIs for screenshots, PDFs, scraping, search and crawling.
For agents it offers a hosted MCP server, an Agent Run API (beta) that runs a managed agent from a plain-English task with a navigation allowlist and step cap, and an Agent Skill Catalog of 136 site-specific skills callable through MCP. Stealth, residential and datacenter proxies, automatic CAPTCHA solving, persisted sessions, authenticated profiles and a 1Password integration keep sessions working on protected and logged-in sites. Hybrid automation pauses a script and hands the live browser to a person, and Session Replay, request logs and credential audit logs record what happened.
The cloud runs from San Francisco, London and Amsterdam endpoints, and Enterprise adds private deployments or licensed self-hosting in a VPC, on premises or air-gapped, with SSO and user roles. The Trust Center lists SOC 2 Type II, GDPR and a BAA for Enterprise. Plans run from a free tier of 1,000 units a month to Prototyping at $35, Starter at $200 and Scale at $500 a month, with units drawn by browser time, proxy data and CAPTCHA solves.
Vendor details
Canonical URL
https://www.browserless.io
Category
Agent infrastructure
Company status
independent
Use cases & customers
Target customers
Deployment options
Integrations
REST APIs with an OpenAPI specification, BrowserQL, Browsers as a Service over WebSocket for Puppeteer, Playwright and Selenium, the Agent Run API (beta), a CLI and a hosted MCP server with a 136-skill catalog; integrations with Browser Use, Stagehand, LangChain, the Claude Agent SDK, Claude Code, Hermes Agent, AgentKit, Anthropic, OpenAI and Gemini computer use, n8n, Make and Zapier; 1Password for credentials.
In practice
You already wrote the Playwright or Puppeteer automation and simply want someone else to run the browsers. Point the existing code at a connection string instead of rewriting it for an agent framework.
Your workload cannot send data to a third party cloud. The same API runs inside your own VPC, on premises or on an air gapped network, with air gapped licensing and an enterprise Docker and Kubernetes deployment guide.
Your agent needs to survive a long task without reloading everything. Persisted sessions keep cookies, cache and local storage between connections, and reconnects keep the browser warm so the automation picks up where it left off.
Sources & related URLs
Agentic Index coverage score
8.5 / 14 capabilities · 61%
| Integrations & Tool Calling | Full |
|---|---|
|
The AI integrations section gives agents a Browserless browser inside Browser Use, Stagehand, LangChain, the Claude Agent SDK, Claude Code, Hermes Agent, AgentKit and computer use agents from Anthropic, OpenAI and Gemini, plus n8n, Make and Zapier. Authenticated profiles and a 1Password integration resolve credentials into sessions, so agents can act while signed in to outside sites. Sourcedocs.browserless.io/ai-integrations/startread 2026-09-23 |
|
| Workflow Orchestration | Partial |
|
BrowserQL scripts chain navigation, waits, conditional logic and structured extraction in one request, and the Agent Run API (beta) runs a managed agent that plans and carries out its own browser steps asynchronously, capped by maxSteps. Both run inside one browser or one agent's loop. There is no workflow composed across steps or agents, and no execution with several participants. Sourcedocs.browserless.io/browserql/startread 2026-09-23 |
|
| Knowledge Grounding & RAG | Not documented |
|
There is no index, vector store or grounding layer over the customer's own content. Smart scrape, search, map and crawl tools turn public web pages into structured output ready for an LLM in the customer's own pipeline, but extraction and search over the public web is not a retrieval structure Browserless maintains over the customer's corpus. Sourcedocs.browserless.io/rest-apis/smart-scraperead 2026-09-23 |
|
| Human Oversight & Guardrails | Full |
|
A person can take over in the middle of a run, and the managed agent works inside set limits. Hybrid automation pauses an automated script, hands control to a person through a secure live URL (interactive or view only), and resumes the automation when they finish, for logins, 2FA, CAPTCHAs or any step that needs human judgment. The Agent Run API constrains a managed agent with a navigation allowlist (allowedDomains) and a step cap (maxSteps), which limit what the agent may do. Sourcedocs.browserless.io/baas/monitor-sessions/hybrid-automationread 2026-09-23 |
|
| Security, Identity & Governance | Full |
|
Browserless holds SOC 2 Type II, is GDPR compliant, and offers a BAA on Enterprise and a DPA for deployments it operates. The SOC 2 scope covers controls across the organization, with infrastructure hosting controls scoped to its cloud, while a self hosted deployment sits inside the customer's own audit boundary. SSO and user roles come with Enterprise, and self hosted deployments have role based access with tokens for each team in admin, developer, viewer and public roles. Sourcebrowserless.io/trustread 2026-09-23 |
|
| Observability & Auditability | Full |
|
The agent's actions are recorded and kept for a period that depends on the plan. Session Replay captures each session as an interactive replay in the dashboard (DOM mutations, mouse movements, clicks, scrolling, keyboard input and console output), request logs can be searched in the dashboard, and each poll of an Agent Run returns the steps the agent has taken with short rationales. The 1Password integration keeps an audit log of every credential resolve, with the reference, origin and outcome but never the value. Replay storage runs from 1 day on Free to 90 days on Scale and custom on Enterprise, and the log search window from 8 hours to 7 days. Sourcedocs.browserless.io/baas/monitor-sessions/session-replayread 2026-09-23 |
|
| Memory & State Persistence | Not documented |
|
What persists is browser state, and there is no agent memory with a stated scope and lifetime. Persisted sessions keep cookies, cache and local storage between sessions, authenticated profiles store logins by name, reconnects keep a browser warm so an automation picks up where it left off, and the MCP browser agent holds one browser session across tool calls. Each is browser state loaded or kept for the next connection, not a memory the agent reads to decide, so it is a setting that helps make the hosted browser ready for production. Sourcedocs.browserless.io/baas/session-management/persisting-stateread 2026-09-23 |
|
| Deployment & Data Residency | Full |
|
Customers choose between their own environment and a list of cloud regions. The same APIs run inside the customer's own VPC, on premises or on an air gapped network, with an enterprise Docker and Kubernetes deployment guide and air gapped licensing. San Francisco, London and Amsterdam cloud endpoints come with every plan, and Enterprise adds private deployments, custom datacenter locations and further locations including the EU. Sourcebrowserless.io/platform/self-hostedread 2026-09-23 |
|
| Prebuilt Agents, Templates & Packs | Full |
|
The Agent Skill Catalog lists 136 skills tested in production across 123 sites and 14 categories (real estate, ecommerce, finance, government, jobs and others). Each is a whole task, such as searching AbeBooks by ISBN and extracting sorted results or finding recent electrical building permits, and each is versioned, listed automatically and callable through the Browserless MCP with structured JSON output. Sourcebrowserless.io/skillsread 2026-09-23 |
|
| Triggers & Channel Coverage | Not documented |
|
Work starts when a caller connects over WebSocket or calls a REST, BrowserQL, MCP or Agent Run endpoint, and no event, schedule or inbound channel of Browserless's own starts agent work. Agent Run's webhooks are outbound notices that a run finished, and the scheduled examples run on Zapier's scheduler. The broad protocol coverage is how callers reach Browserless, not what starts it. Sourcedocs.browserless.io/overview/connection-urlsread 2026-09-23 |
|
| Model Flexibility & Routing | Not documented |
|
The one model Browserless runs, behind the Agent Run API (beta), is fully managed. The customer supplies no model key, and there is no model or provider selection. Everywhere else the model lives in the customer's own agent (Browser Use, Stagehand, LangChain, computer use agents), and choosing which assistant calls Browserless is not the same as choosing a model inside it. Sourcedocs.browserless.io/rest-apis/agent-runread 2026-09-23 |
|
| APIs, SDKs & MCP Extensibility | Full |
|
Outside systems can call Browserless through REST APIs with a published OpenAPI specification, BrowserQL (an automation language built on GraphQL, with an editor), Browsers as a Service over WebSocket for Puppeteer, Playwright and Selenium, the Agent Run API, a CLI, and a hosted MCP server with scrape, search, map, crawl, function and agent tools. Sourcedocs.browserless.io/open-apiread 2026-09-23 |
|
| Testing, Debugging & Optimization | Not documented |
|
There are no fixtures, scoring, quality gates or comparison of agent runs. The MCP server's performance tool runs Lighthouse audits that score web pages, not the customer's agent workflows, and Session Replay and hybrid automation record and hand over sessions instead of evaluating them. Sourcedocs.browserless.io/mcp/browserless-mcp-serverread 2026-09-23 |
|
| Browser & Computer Use | Full |
|
Operating a real browser for agents is the product. Browserless runs managed Chrome, Firefox and WebKit that an agent drives through Puppeteer, Playwright, Selenium, BrowserQL, the MCP browser agent or computer use agents from Anthropic, OpenAI and Gemini, with stealth, residential proxies, automatic CAPTCHA solving and Chrome extensions for production use. Sourcebrowserless.ioread 2026-09-23 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Product updates
Version 1.37.0 of the Browserless MCP server adds a tool that creates, checks or closes a live viewer link for a browser that is already running. Status calls report whether the link is open, closed or expired and how many people are viewing it, and new links default to view only.
Bears on: Human approval / guardrails
View sourcePricing
Free 1,000 units a month · Prototyping $35/mo ($25 billed annually) plus overage
Units: 1 unit per 30 seconds of browser time per connection (reconnects start a new unit), 6 units per MB of residential proxy, 2 per MB of datacenter proxy, 10 per successful CAPTCHA solve; plan units included, overage per unit
What is public
The pricing page publishes monthly and annual plan prices, the free tier, unit definitions, overage rates, concurrency and session limits, replay storage by plan and the Enterprise feature list, and compliance posture is published in a trust center.
Billing mechanics
A free tier of 1,000 units a month, then Prototyping at $35, Starter at $200 and Scale at $500 a month, or $25, $140 and $350 billed annually, about 30 percent less. A unit is up to 30 seconds of browser time; proxies and CAPTCHA solves draw extra units, and overage runs $0.0020 to $0.0015 per unit. Enterprise is custom and adds private infrastructure, GPU support, single sign on and a dedicated technical account manager.
Cost watchouts
Every reconnect starts a new unit, residential proxy data costs 6 units per MB and each CAPTCHA solve 10 units, so proxy-heavy or reconnect-heavy agents use units far faster than browser time alone suggests; Session Replay is not on the free plan.
Variable cost rationale
Units are drawn by browser time, proxy bandwidth, CAPTCHA solves and reconnects, so session length, proxy use and anti-bot needs drive cost above the plan's included units.
Additional watchouts
Units are browser time, so long sessions, proxy data and CAPTCHA solves use them faster than a request count suggests, and concurrency is capped by plan, so a workload that fans out browsers in parallel can force a higher tier. The self hosting license distinction also catches teams out: free for non commercial use, licensed for commercial.
Overage / add-ons
Overage rates plus separate unit consumption for proxies and CAPTCHA solving are billed in addition to plan allowances
Sales call required
Mixed (some tiers require a call)
Free / trial
Free plan at $0 with 1,000 units a month, no card required
Lowest paid plan
Prototyping $35 a month ($25 billed annually) plus overage
Commercial notes
Being bootstrapped rather than venture funded is commercially relevant here: the pricing has been stable and incremental for years rather than repriced around a growth thesis. Teams with existing Puppeteer and Playwright code connect it over WebSocket without a rewrite. The self hosted route also caps vendor spend entirely for teams willing to run the browser pool themselves, subject to the commercial license.
Key ambiguities
Cost depends on the mix of browser time, proxy bandwidth, CAPTCHA solves and reconnects, all drawn from one unit pool; Enterprise and self-hosted licensing are custom.
Missing data
Not established: how many units proxy data and a CAPTCHA solve draw, the cost of a commercial self hosting license, and Enterprise pricing; ask the vendor.
Related vendors
- AgentOps — Agent observability and debugging platform: open source SDKs trace…
- AgentPhone — AgentPhone gives the AI agents a team builds their own US or…
- Agno — Python agent framework and AgentOS runtime (formerly Phidata) for…
- AIsa — Resource and payment gateway for AI agents, with one key to 110+…
- Albato — Albato is a no code integration platform that sells an embedded,…
- AlphaBitCore — AI control plane for regulated financial firms: one gateway enforces…
Alternatives to Browserless
The closest documented capability profiles to Browserless among agent infrastructure platforms tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Kernel8.5 / 14Adds documented Testing, Debugging & Optimization
- Steel7.0 / 14Adds documented Testing, Debugging & Optimization
- Anchor Browser11.0 / 14Adds documented Triggers & Channel Coverage and Model Flexibility & Routing, among others
- Hyperbrowser8.0 / 14Adds documented Memory & State Persistence and Model Flexibility & Routing
- Lakera7.0 / 14Adds documented Testing, Debugging & Optimization
- OpenBox AI6.0 / 14Adds documented Testing, Debugging & Optimization
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded