Your agent reads a snippet, not the page
Most fetch tools do a static request and summarize through a small model: numbers rounded, caveats dropped, tables flattened. Gourdo returns the actual rendered page, so the agent reads the source.
Headless browser API · MCP connector live
Give your AI agent a real browser.
Gourdo renders any web page, JavaScript and all, through rotating residential IPs, and returns what a person would see. Markdown, HTML, screenshots, PDF. One URL to connect Claude, Cursor, ChatGPT or your own pipeline.
$5/month + data metered to the byte · 14-day free trial with a data allowance included
Gourdo… gourd… pumpkin… pumpkin head… headless horseman… headless browser. You get it.
Connect the AI clients over MCP; call it from workflow tools and frameworks with one HTTP request.
Why Gourdo
Your agent is reading a snippet. Give it the page.
Half the web doesn't render for a fetch
Client-rendered apps, sites that block datacenter traffic, pages that load after "done". A real browser on a residential IP gets what a person gets.
Safer than handing over your own Chrome
Gourdo's browser holds none of your logins, cookies or sessions, runs nowhere near your machine, and is read-only. Worst case: it reads a public page and bills a few cents.
MCP connector
Connect in three steps. No code.
Works with claude.ai, Claude Desktop, Cowork, Claude Code, Cursor, ChatGPT, VS Code, Gemini CLI, or any MCP client.
- 1Create an API key with “Allow MCP connections” ticked.
- 2Add the connector in your client:
https://gourdo.ai/api/mcp - 3Paste the key when the sign-in page opens. Done. Your agent has Gourdo’s tools.
What the agent gets · 13 tools
- scrape_markdown
- scrape_html
- extract_elements
- get_page_links
- get_page_metadata
- take_screenshot
- scrape_pdf
- create_session
- get_session
- list_sessions
- close_session
- close_all_sessions
- get_usage
Scrape as Markdown, HTML, elements, links or metadata. Screenshots the agent can look at. PDF. Session control. And get_usage: your balance and account state in plain language, so the agent can tell you what to do if a call is refused.
Scope and revoke
The key’s permissions decide what the agent may do. Deactivate or delete the key and the agent is disconnected on its next call.
Read-only today: the agent reads pages, it doesn’t click, type or log in. Browser automation and a secrets store are on the roadmap.
How it works
Request in. Rendered page out. Every byte counted.
- 01 · RequestYour agent or script sends a URLOver MCP or one HTTP call. Pick the output format and, if needed, a wait strategy.
- 02 · RenderA real Chromium page opensIn Gourdo's managed browser pool. JavaScript runs. The page loads the way it would for a visitor.
- 03 · EgressTraffic leaves through a residential IPRotating on every request. Every byte on the wire, both directions, is metered.
- 04 · ResultYou get the format you asked forWith a header: final URL, status, bytes, session id. Sessions carry cookies across pages and close when you say so, or after 5 idle minutes.
Every call tells you what it cost
The first line of every result is the receipt. The same numbers show on your dashboard as they happen.
Scraped https://example.com/pricing — HTTP 200, 1904 ms, 412380 bytes metered. sessionId: 7f3c…e21a
What you can do
Outcomes first. Just enough tech.
- Read the real pageClean Markdown of the main content, or the full rendered HTML.
- Pull one pieceElements by CSS selector; links for site mapping; metadata and Open Graph for identifying a page.
- See itViewport or full-page screenshots the agent can look at; PDF for a dated record.
- Get throughFull JavaScript rendering, rotating residential IPs, stealth browser, wait strategies for pages that never go idle.
- Reuse the sessionCookies carry over; faster on the second page; close when done.
- Know what it costMetered to the byte, drawn from a prepaid balance that is also your hard spending cap. No surprise invoices.
Use cases
What people point it at.
Data aggregation pipelines
Many businesses depend on pulling fresh data from a set of sites on a schedule. Gourdo is one HTTP call per page from n8n, Make, Zapier, LangChain, CrewAI or your own agent: aggregate, enrich a CRM, ingest into a vector store, crawl via links.
See a worked example →Research with sources
Search finds the URLs; Gourdo reads them in full. Cite the page, not a summary of it.
Seeing, not just reading
Charts, maps, infographics, "does it still show what I saw", answered with an image.
Watch pages on a schedule
Scheduled agents scrape, compare to last run, and act through your other connectors: a task, a CRM update, an email.
Deliverables and evidence
Render to PDF: terms, notices, confirmations, prices, as they stood on the day.
Check your own site
Screenshots at several viewports after a deploy, OG tags, and your site as a residential visitor sees it.
Pricing
One plan. A flat fee plus the data you actually use.
Your balance is your spending limit. Scraping stops at zero unless you opt into overflow credit.
Worked exampleReading 50 pages at 1–2 MB each ≈ $0.20–$0.40 of data.
Beta pricing while we’re early. Rates may change when the beta ends.
Gourdo$5/month
Data, prepaid, metered to the byte$4 / GB
Optional overflow credit when your balance hits zero$8 / GB
- 14-day free trial on the platform fee, with a data allowance to try it on
- Rotating residential IPs on every request
- MCP connector for your agents
- Live usage dashboard; your prepaid balance is the cap
- Cancel anytime
Prices shown live from billing. Billing FAQ
FAQ
The questions people ask before the first call.
Billing, sessions, refusals, and the limits an agent should know.
Read the full FAQWhat am I paying for?
Two things: a flat monthly subscription that buys access, and the data you use, paid from a prepaid balance you top up. The subscription is never taken out of your balance.
What exactly is measured?
Data transferred over the network, both directions, for everything your request causes: page, images, scripts, fonts, pixels. Not per request.
Do failed requests cost money?
Yes. A timed-out, blocked or error page still moved data. Don't retry in a loop against a site that is blocking you.
Can I be charged more than I funded?
Your balance is your spending limit; scraping stops at zero. Overflow credit is off by default and capped. The cap is where scraping stops, not an exact maximum: in-flight requests finish and may land a few cents over.
How does the trial work?
The subscription fee is waived for 14 days and you get a data allowance to evaluate with. When it ends, your prepaid balance pays for data. Fund it before then to avoid a gap.
Can my agent log in to sites or click things?
Not yet. Sessions are read-only: Gourdo renders and reads pages. Browser automation and a credentials store are on the roadmap.
Is it safe to give an agent?
Gourdo's browser holds none of your logins or cookies, runs on our infrastructure, and cannot act on a page. The key you create scopes what the agent may do, and revoking it disconnects the agent. Treat scraped text as data, not instructions.
Why did my balance drop after I stopped?
An open session keeps a browser whose background traffic moves a little data until it expires (5 idle minutes). Close sessions when done. It's the single most effective way to cut cost.
Give your agent the whole web.
Beta pricing while we’re early. A data allowance included with the trial.
Get startedQuestions? hello@uelabs.ai