Connector

The Connector is a separate MCP endpoint for AI assistants and app-directory listings. It gives an assistant a cloud browser to navigate pages, read, fill forms, capture screenshots, run Lighthouse audits, search the web, and export pages.

For broader browser automation tasks, use the full MCP server instead.

Prerequisites

• A Browserless account
• An MCP-compatible client (Claude, ChatGPT, etc.)

Endpoint

https://mcp.browserless.io/mcp/connector

The full server remains available at https://mcp.browserless.io/mcp.

Authentication

Authentication is identical to the full server. OAuth discovery is served at the domain root and shared by both endpoints, so signing in once works for either.

Method	Best for
OAuth (Browserless account login)	Clients that support OAuth — no token needed
Authorization header	Clients that support custom headers

OAuth

Sign in with your Browserless account when the client prompts you. No token to copy or manage. OAuth is enabled on the Connector at https://mcp.browserless.io/mcp/connector with no extra configuration.

API Token

Copy your token from the account dashboard, then send it in the Authorization header:

Authorization: Bearer YOUR_API_TOKEN_HERE

Tools

The Connector exposes five tools.

browserless_agent

Drives a real browser across multiple steps (navigate, read, click, type, screenshot, work across tabs) to complete a task you direct. See the Browser Agent page for the full command reference.

Example prompts:

Go to https://example.com and tell me the page heading and where its link goes.

Open the docs page at <url>, then take a screenshot of the page.

browserless_search

Searches the web and returns web, news, or image results.

Parameter	Type	Required	Default	Description
query	string	Yes	—	The search query
limit	number	No	10	Maximum results (max 100)
sources	string[]	No	["web"]	web, news, images
tbs	string	No	—	Time filter: day, week, month, year
lang / country / location	string	No	—	Locale / geo hints

Example prompt:

Search the web for the latest documentation on the Model Context Protocol.

browserless_export

Fetches a URL and returns its content in the native format (HTML, PDF, or image), auto-detecting the content type.

Parameter	Type	Required	Description
url	string	Yes	The URL to export (http or https)
gotoOptions	object	No	Navigation options (waitUntil, timeout, referer)
bestAttempt	boolean	No	Proceed even if awaited events time out
waitForTimeout	number	No	Milliseconds to wait after load before exporting

Example prompt:

Export the page at https://example.com.

browserless_performance

Runs a Lighthouse audit on a URL and returns scores and metrics.

Parameter	Type	Required	Description
url	string	Yes	The URL to audit (http or https)
categories	string[]	No	accessibility, best-practices, performance, pwa, seo — omit for all
budgets	object[]	No	Lighthouse performance budgets

Example prompt:

Run a Lighthouse performance audit on https://example.com and list the top issues.

browserless_skill

Loads an on-demand agent skill: concise guidance the agent uses when a page needs special handling. The agent also auto-loads the relevant skill mid-task; this tool is the explicit fallback.

Parameter	Type	Required	Description
id	string	Yes	One of: shadow-dom, cookie-consent, modals, snapshot-misses, dynamic-content, screenshots, tabs

Scope

The Connector focuses on user-directed automation on content you point it at. Anything beyond that — single-shot scraping, site crawling, sitemap discovery, custom code, or proxy control — lives on the full MCP server. Not part of the Connector:

• Tools: browserless_smartscraper, browserless_crawl, browserless_map, browserless_function, and browserless_profiles.
• Agent commands: arbitrary code execution (evaluate), CAPTCHA solving, credential/profile loading, and file transfer (uploadFile / getDownloads).
• Parameters: proxy routing, authentication profile / createProfile (agent, export, performance), bulk-asset includeResources (export), and per-result scrapeOptions (search).

The Connector operates only on content you direct it to. It does not bypass access controls or bot protection, solve CAPTCHAs, or route around IP/geo restrictions.