Platform
Five tools for your agent. One account.
Scrape, crawl, search, real browsers and agents, all paid in the same credits, from 1 credit a page. Connect Claude Code or Codex once, then ask in plain words.
1,000 free credits every month. No card needed.
Products
Pick the tool for the job. Or let your agent pick it.
Scrape
Any web page as clean markdown or JSON your model can read, including pages that only load with JavaScript. For example, a product page into its name, price and stock level.
1 credit a page · about $0.0032 to $0.0049
See scrapeCrawl
One link in, every page on the site back, each sent to you as soon as it is read. For example, a company’s whole docs site into your search index.
1 credit a page · about $0.0032 to $0.0049
See crawlSearch
The top results for a query, with each page already read if you ask, so your agent answers in one step. For example, this week’s coverage of a topic, with the article text.
5 credits per 10 results · about $0.016 to $0.025
See searchBrowsers
A real browser in the cloud that your agent drives with Playwright, Puppeteer or our TypeScript SDK. Logins carry over between sessions, and every session is recorded.
1 credit a minute · about $0.19 to $0.29 an hour
See browsersAgents
Describe a task in plain words. An agent does it in a real browser, reports what it found, and you can take over at any point.
About 15 credits a minute · about $0.07 to $0.08
See agents
Ways in
Connect once, then ask. Your agent calls the tools for you.
claude mcp add --transport http outcrawl https://api.outcrawl.ai/mcpYour AI agent
Add Outcrawl to Claude Code, Codex or Cursor with one command, then ask in plain words. Your agent reaches 27 tools and picks the right one. The first call opens our sign-in page, and signing in creates your account.
How the MCP server worksYour terminal
One command-line program, for agents that work in a shell and for jobs that run on a schedule. It reaches 55 of the 60 calls in the API.
How the CLI worksYour own app
When your agent builds Outcrawl into an app, it uses the TypeScript or Python SDK, which reach all 60 calls. The TypeScript SDK also opens a live browser.
How the SDKs work
What each reaches
Most of the platform, from all three. Here is exactly where they differ.
Your agent's crawl call answers with every page once the crawl ends. Checking on a crawl, paging through its pages and stopping it are in the CLI and the SDKs.
A live browser stays connected for the whole session, which a tool call or a command cannot. Your agent writes a short Playwright or Puppeteer script against one URL instead, or hands the task to an agent.
Your agent can list sessions, open one and read what went wrong. Full event and network logs, videos, HAR files and sharing are in the CLI and the SDKs; exporting a session is in the SDKs.
A person stores the passwords, cards and billing profiles agent runs may use from the CLI or the SDKs. The console lists and revokes them: passwords under Secrets, cards and billing profiles in Wallet. Not over MCP, on purpose: your agent reads web pages, and a page could ask it to.
Connecting a card provider, so each run gets a card of its own, reading every purchase and setting when one waits for a yes are done by a person, in the console, the CLI or the SDKs. Not over MCP, on purpose: a page your agent reads could ask it to connect a card or turn that question off.
Connecting another MCP server, such as Gmail or Slack, for agent runs to use is done by a person, in the console, the CLI or the SDKs.
FAQ
Questions about the platform.
outcrawl_agent. Saved logins and session recordings are MCP tools either way.oc.browser(), is TypeScript only; from Python, Playwright connects to the same URL.