Sign inGet started

Public web data for your AI pipelines

Collect training sets, RAG sources and agent context through residential and mobile IPs in 249 countries and territories. Pay per GB, with no hidden concurrency limits.

Get startedView Docs
Many data sources wired through one gateway into an AI model

Why AI data collection gets blocked

Crawlers and agents run in a datacenter. Many sites answer those IPs with a block page, and it still says 200 OK.

From a datacenter
$ curl -i https://shop.example/product/42
HTTP/2 200
content-type: text/html
<title>Access denied</title>
<h1>Please verify you are a human</h1>
Fails: The pipeline stores the block page as content, and the model learns from it.
Through 2extract
$ curl -i -x http://2xt-customer-…@proxy.2extract.net:5555 \
https://shop.example/product/42
HTTP/2 200
content-type: text/html
<title>Trail running shoe</title>
<span class="price">$89.00</span>
Works: The same request from a residential IP gets the real page: about 99% of requests through 2extract succeed.

Let the agent set up its own proxy

Connect the MCP server once. The agent creates a capped residential proxy, routes its requests through it and reports what it collected.

~/projects/price-tracker

Example session. Account ID, password and IP are illustrative.

  1. 1Get an API key. Sign up, top up from $10 and create a key under API keys in the dashboard.
  2. 2Connect. One command adds the 2extract MCP server to Claude Code, Cursor or any MCP client.
  3. 3Describe the job. Country, city, session and budget, in plain language.
  4. 4The agent builds it. It finds the city, creates the proxy and sets the cap.
  5. 5It ships the code. The connection string lands in your scraper, and the first request runs.
Set up the MCP server (opens in a new tab)View full documentation

What a data pipeline needs from the network

Everything is set in the username or through the API, so the pipeline controls it per request.

What a data pipeline needs from the network

  • Residential and mobile IPs

    Real-user IPs that sites serve the full page to, in 249 countries and territories.

  • Rotation or sessions

    A fresh IP on every request for broad crawls; a sticky session for flows that span several requests.

  • Targeting in the username

    Country, region, city, ZIP or ASN, so one pipeline collects every market it needs.

  • Parallel at scale

    No hidden concurrency limits: run as many workers as the job needs.

  • A budget per pipeline

    One proxy per dataset or agent, capped daily or monthly in GB or USD, with a per-day usage export.

  • Agents that set themselves up

    Through the MCP server, Claude Code, Cursor or any MCP client creates and caps its own proxy.

Pay per GB from $3.99, top up from $10, no commitment

Get startedSee pricing

AI data collection questions

Targeting, sessions and limits are in the documentation.

Still have a question?

Talk directly to the engineers who built the product. No chatbots, no first-level support scripts.

Your pipeline's first request is one line away

Sign up with Google, GitHub or email, top up from $10 and pay per GB. No commitment.

curl -x http://…-country-us:•••@proxy.2extract.net:5555 ipinfo.io/json
Get startedIntegration guides