scrape1 creditFetch one URL. Returns markdown, JSON, HTML, or a screenshot. Sync, under 60 seconds.
AI Robot Factory — paste a URL, get a scraper in 30 seconds
Paste a URL.
Your AI agent builds the scraper.
In 30 seconds.
Hosted 24/7, rescraped on a schedule, exposed as a live REST API and an MCP tool. Zero selectors, zero code, zero babysitting.
Live API3 of 3 free runs left today. Sign up for unlimited.
Idle
Hit Run scrape to send a real request to /v1/scrape. You'll get a run id, the credit cost, and the actual response.
Built by people who ship
A product in daily use.
Early-access teams are scraping tens of thousands of URLs a week — no selectors written, no cron jobs babysat.
“Built a Google Maps lead monitor in six minutes. That would have been a week of dev and a weekend of debugging before.”
Marco D.beta user
Growth Lead at a YC B2B SaaS
“I paste a URL on Monday morning. By Tuesday I have fresh rows in Sheets. I stopped writing selectors three weeks ago.”
Ana R.early access
Ops Manager, 40-person e-commerce
“The MCP tool means Claude keeps our competitor pricing warm on its own. We shut down the dedicated scrape stack.”
Jonas K.early access
CTO, 15-person startup
Powered by the same stack as your favourite apps
- Neon
- Stripe
- Clerk
- Vercel
- Upstash
How it works
From URL to production scraper, in three acts.
The AI agent does the selector writing, the pagination heuristics, and the re-run loop. You bring the URL and keep the data.
Paste a URL
Drop the page you want to extract — a product listing, a directory, a hub. Optionally describe what you need, or just let the agent figure it out.
30-second promise
The AI agent builds your scraper
It navigates in a cloud browser, detects the main list, infers columns, picks a row key, and compiles a versioned Robot Program. You see every step streamed live.
Zero selectors, zero code
Use it anywhere — forever
The robot runs 24/7 on your schedule, detects changes between runs, and exposes its data through a live REST endpoint, an MCP tool, and webhooks. When the site breaks, the agent self-heals.
Dashboard · API · MCP · Webhooks
One product, three surfaces
Use Stekpad from where you already work.
Five verbs. Predictable JSON. Real run ids.
POST a URL, get markdown, JSON, HTML or a screenshot. Async crawls return a run id you can poll or webhook. SDKs in TypeScript and Python.
- scrape · crawl · map · extract · search
- Idempotent run ids you can poll or webhook
- TypeScript + Python SDKs, MIT-licensed
- Predictable per-verb credit cost
Why Stekpad
Four things you only get here.
App-native storage and enrichment
Every scrape lands in a persistent, editable dataset. Re-run it next week, add a column, filter, export. Other APIs give you a payload and forget it. Stekpad gives you a payload and a place to keep it.
19 native enrichers, zero third-party vendors
Find emails, verify them, look up companies in OpenCorporates and Companies House, geocode addresses, classify text with an LLM. We run all 19 ourselves. Your scraped data never leaves our stack.
Cookie bridge, zero stored cookies
Need a logged-in page? The Chrome extension fetches it inside your own browser and posts back the rendered HTML. Your session cookies never leave your machine. Not even encrypted, not even briefly.
First-class MCP server
Eight MCP tools shipped on day one: scrape, crawl, map, extract, search, list_datasets, get_dataset, query_dataset. Reads are free. Writes charge credits. Errors are typed. Your agent never blocks for more than 60 seconds.
Under the hood
Your robot sits on top of a real API.
Every capability the AI agent uses is a documented verb you can call yourself. Skip the robot when you want raw control — use the robot when you want it compiled, hosted, and kept alive.
crawl1 / pageWalk a site. Honor robots, set include/exclude paths, dedupe by canonical URL. Async, with webhooks.
map1 / 1k URLsDiscover URLs from sitemaps, robots, and a shallow link walk. Fast, no rendering. Use it before a targeted crawl.
extract5 / URLPass a JSON schema and a prompt. Get back structured fields, validated, retried twice on schema failure.
search5 + 1 / resultRun a web search. Optionally scrape every result in the same call. Built for agents.
Most users never touch these directly — the AI agent does.
Skip the API, let the agent build your scraper →Your data never leaves our stack
19 enrichers in the box.
One credit per row, except linkedin_enrich (premium, v1.5).
ai_extractLLMai_summaryLLMai_classifyLLMai_translateLLMai_cleanLLMai_embedLLMfind_emailsWebfind_phonesWebfind_socialsWebfind_tech_stackWebfind_company_infoWebfind_faviconWebemail_verifyValidationwhoisValidationdnsValidationphone_enrichSpecializedcompany_enrichSpecializedemail_finderSpecializedgeocodeSpecializedAuthenticated scraping, the right way
Your session cookies never leave your browser.
- 01
You call /v1/scrape
Fire a normal request with
use_session: "stripe.com". - 02
We push a fetch job
Our backend pushes the job over a workspace-scoped websocket to your Chrome extension. No cookies, no auth tokens — just the URL and the actions.
- 03
Your browser fetches the page
It attaches its own cookies, posts back the rendered HTML. We store the content. We never see the cookies.
Cookies never touch Stekpad servers. Architecturally. Not as a policy — as a wiring diagram.
vs Firecrawl
Same shape. Different product.
| Feature | Stekpad | Firecrawl |
|---|---|---|
| Built-in dataset storage | Yes, every scrape | No, returns by value |
| Native enrichers | 19, all in-house | 0 |
| Authenticated scraping | Cookie bridge, zero stored cookies | Server-side cookies |
| MCP server | First-class, 8 tools | Community |
| Pricing model | PAYG packs, no subscription required | Subscription |
| Credit cost / scrape | 1 | 1 |
Built for AI agents
Works with Claude, Cursor, and Claude Code.
Stekpad ships an MCP server as `npm install -g @stekpad/mcp` and as a hosted HTTP MCP endpoint for Claude Desktop. Reads are free. Writes charge credits from your workspace wallet. Every tool call returns a credits_charged so your agent can budget itself.
claude_desktop_config.json
json
{ "mcpServers": { "stekpad": { "command": "npx", "args": ["-y", "@stekpad/mcp"], "env": { "STEKPAD_API_KEY": "stkpd_live_..." } } }}FAQ
Questions we get asked a lot.
How is Stekpad different from Firecrawl or Browse.ai?
Firecrawl is an API — you still write the extraction schema. Browse.ai asks you to record every click. Stekpad does neither: an AI agent opens the page, detects the list, infers the columns, and compiles the scraper. You paste a URL, you get a working dataset in about 30 seconds. It then runs on a schedule, with delta detection, no cron to babysit.
Does it work on pages behind a login?
Yes. The Chrome extension acts as a cookie bridge — it fetches the page from inside your own browser and posts back the rendered HTML. Your session cookies never leave your machine, encrypted or otherwise. A managed credential vault is on the roadmap for teams that want to skip the extension entirely.
What happens when the site changes its layout?
The run does a self-healing pass. If the extractor can't find the expected columns, the AI agent re-plans against the new DOM and patches the robot program. You get an email if it had to change something, and a diff of the old vs new plan in the History tab. No manual fixing in most cases — sites that fully redesign may still need a human nudge.
Is web scraping legal?
For public data, it's broadly fine in most jurisdictions — hiQ v. LinkedIn in the US is the usual reference. That said: respect
robots.txt, stay within a site's terms of service, rate-limit politely, and never scrape personal data you aren't allowed to process under GDPR/CCPA. We have a longer write-up in the docs.What happens at the end of my free tier?
Nothing dramatic — 300 credits reset on the 1st of every month, forever. No card on file, no surprise charges, no auto-upgrade. If you run out mid-month, the robot pauses and waits for the reset, or you can top up with a PAYG credit pack.
Can I self-host?
Not today. Stekpad is cloud-hosted so the browser pool, queue, and MCP gateway are managed for you. A self-hosted tier for regulated industries is on the roadmap, but it's not something we'll ship in v1.
How does the MCP server work?
Every workspace gets a dynamic Gateway at
mcp.stekpad.com/{api_key}. The gateway introspects your robots and exposes each one as a typed MCP tool — get_competitor_prices, list_new_yc_companies, and so on. Paste the URL into Claude Desktop, Cursor, or Claude Code once and every future robot shows up automatically.Can I cancel anytime?
There's nothing to cancel — Stekpad is pay-as-you-go. Buy a credit pack when you need one, stop using the product when you don't. Starter and Pro are month-to-month and can be downgraded to free with one click, no phone call, no retention flow.
Get an API key. Scrape something. Keep it.
300 credits a month, free, forever. No card. No subscription. The MCP server is free to read your own data.