outcrawl

CLI

The web, from a shell. Run by your agent.

For Claude Code, Codex and any agent that runs commands, and for the jobs it schedules. Read pages, crawl sites and run agents at 1 credit a page, about $0.0032 to $0.0049.

Install the CLIRead the CLI docs

1,000 free credits every month. No card needed.

Terminal
The command: outcrawl crawl https://docs.astral.sh/uv/ --limit 10
Outputstdout
exit 00.0 s10 credits · $0.049

Recorded from outcrawl 0.4.8 on 8 October 2026, as an agent runs it.

Install

One line, then sign in. Your browser opens. No key to paste.

  1. 01

    Install it.

    One line on macOS or Linux, Homebrew, or Scoop on Windows. One program, nothing else to install first.

  2. 02

    Sign in.

    outcrawl login opens your browser with a code. Approve it and the CLI stays signed in. If you are new, signing in creates your account with 1,000 free credits.

  3. 03

    Tell your agent.

    Claude Code and Codex run commands in your terminal, so they can use it as soon as it is installed. Ask for what you want and say “use outcrawl”.

terminalbash
curl -fsSL https://get.outcrawl.ai/cli/install.sh | sh
export PATH="$HOME/.local/bin:$PATH"

macOS and Linux. The second line puts it on your PATH in this shell.

On a server or in CI, nothing opens a browser: set OUTCRAWL_API_KEY to a key from the console and it takes priority over a stored sign-in.

Commands

55 of the API’s 60 calls. Including the ones MCP leaves out.

Not in MCP

  • Scrape

    1 command

    Read any page as markdown, or as JSON with the fields you name.

    • scrape

  • Crawl

    4 commands

    Read a whole site from one link. Pages stream in as they are read; check on a crawl later, page through it, or stop it.

    • crawl
    • Not in MCP: crawl status
    • Not in MCP: crawl results
    • Not in MCP: crawl cancel

  • Search

    1 command

    Search the web, and read the top results in the same call.

    • search

  • Agents

    11 commands

    Hand off a task that needs clicks, forms or a sign-in. Follow its steps, answer what it asks, read what it found, take over, give it a file, or stop it.

    • agent
    • agent list
    • agent status
    • agent events
    • agent results
    • agent cancel
    • agent control
    • agent answer
    • agent files add
    • Not in MCP: agent secrets add
    • Not in MCP: agent purchases

  • Monitors

    3 commands

    Watch a page on a schedule and get a Slack, Discord or webhook message when it changes.

    • monitors create
    • monitors ls
    • monitors rm

  • Saved logins

    4 commands

    Create, find and delete saved logins, so a read or a run starts signed in.

    • profiles create
    • profiles ls
    • profiles get
    • profiles rm

  • Session recordings

    10 commands

    Every request, console line and event a session recorded, as a HAR file or video links, and who may watch it.

    • sessions list
    • sessions get
    • Not in MCP: sessions events
    • Not in MCP: sessions network
    • sessions digest
    • sessions request
    • Not in MCP: sessions console
    • Not in MCP: sessions har
    • Not in MCP: sessions videos
    • Not in MCP: sessions share

  • Rules

    2 commands

    Read and set the rules every agent run must follow.

    • rules get
    • rules set

  • Usage and credits

    2 commands

    What you have spent, by day or by command, and what is left.

    • usage
    • credits

  • Saved credentials

    4 commands

    Store a password, a card or a billing profile a run can use but never read. A card is kept without its security code. The value is typed hidden or piped in, so it stays out of your shell history.

    • Not in MCP: secrets create
    • Not in MCP: secrets ls
    • Not in MCP: secrets rm
    • Not in MCP: secrets import

  • Wallet

    8 commands

    Connect your own card provider, such as Brex or Stripe Issuing, so each run gets a card of its own; the key is asked for hidden and tested first. See every purchase, and set when one waits for a person’s yes.

    • Not in MCP: purchases ls
    • Not in MCP: wallet settings
    • Not in MCP: wallet settings set
    • Not in MCP: wallet providers
    • Not in MCP: wallet connections
    • Not in MCP: wallet connect
    • Not in MCP: wallet test
    • Not in MCP: wallet disconnect

  • Connected tools

    5 commands

    Connect another MCP server, such as your mail, for agent runs to call. Test it first; nothing is stored until it passes.

    • Not in MCP: integrations connect
    • Not in MCP: integrations test
    • Not in MCP: integrations ls
    • Not in MCP: integrations update
    • Not in MCP: integrations rm

Five calls are SDK only: exporting a session as one bundle, reading one connected card provider, minting an upload token for a saved credential, minting an upload token for one run’s temporary secret and minting an upload token for connecting a card provider.

Output

Output a script can trust. Data on stdout, everything else on stderr.

  • --json

    Exactly the API's answer on stdout, and nothing else: no colour, no hints, no banner.

    any command

  • ndjson

    A crawl, an agent's steps and a session's events and requests stream one JSON object a line, printed the moment each arrives.

    crawl, agent events, sessions events, sessions network

  • stderr

    Errors, notices and the update check. Under --json an error is a JSON object with a code and a message.

    never mixed into stdout

  • exit 0 1 2 3

    0 it worked, 1 the call was made and failed, 2 the command was wrong and nothing was sent, 3 the API refused it: sign-in, plan or workspace.

    $?

Without a terminal it never stops to ask: a missing id is an error with exit 2, not a wait. Colour only appears at a terminal, and NO_COLOR turns it off.

Examples

Ask, and your agent runs it. The commands it reaches for.

Ask your agentDocs

Reference documentation without leaving your editor.

example
Pull in the authentication section of this API's docs before we wire up the client. Bash(outcrawl scrape https://…): the section as markdown, straight into context.

Interactive

Or open it and type. For when you would rather do it yourself.

Run outcrawl on its own at a terminal and it opens a prompt, with your plan and credits left at the top.

  • Type a task and an agent run starts. Its steps appear as it takes them, then its answer and what it cost.

  • /scrape <url> shows a page as text. /search <query> numbers the results, and /scrape 2 reads the second.

  • /crawl <url> lists pages as they arrive. Esc asks before it stops anything.

  • /budget 5 caps the runs you start next at 5 credits. /control gives you the live browser of a running agent.

  • /runs, /usage and /account show your latest runs, this month’s spend, and who is signed in with which keys.

  • Quitting leaves a running agent or crawl going, up to its budget. Stop it with Esc first.

Pricing

Same credits, same rates. Every command says what it cost.

Tool

Dollars on

  • Scrape

    Any web page as clean markdown or JSON.

    1 credit a page

    $4.90 per 1,000 pages

  • Crawl

    Every page on a site, from one link.

    1 credit a page

    $4.90 per 1,000 pages

  • Search

    Web results for your agent. Reading a result page is 1 credit a page on top.

    5 credits per 10 results

    $2.45 per 1,000 results

  • Agents

    A task in plain English, done in a real browser.

    About 15 credits a minute

    about $0.07 a minute

Agent runs are billed for the AI model's usage, browser time and proxy usage, in credits. The estimate is a typical run; the actual cost depends on the task. The middle half of runs land between 10 and 25 credits a minute on Builder. Model usage converts at your plan's price per credit, so on Pro and Scale the same minute is more credits.

  • An agent run you start here stops at 2,000 credits, about $6.39 to $9.80, unless you pass --budget. It keeps what it found.

  • outcrawl credits shows what is left this month. outcrawl usage breaks your spend down by day or by command.

See full pricing

FAQ

Questions about the CLI