Agents don't crash. They believe.
When a scraper hits a challenge page, a cookie wall, or an empty SPA shell, it rarely throws an error. It returns the wrong page as if it were content. Your agent reads it, trusts it, and confidently does the wrong thing. Nobody notices until a customer does.
<h1>Checking your browser
before accessing…</h1>
Your agent thinks this is the pricing table. It will quote it to your customer.
deductions: {"empty_body":0.7,"http_error":0.7}
billed: false
Your agent knows it got blocked, and routes around it. That is the whole product.
$ curl -sA "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/128.0 Safari/537.36" \-o body.html -w "%{http_code}\n" https://www.g2.com/403$ curl -s https://api.enconvert.com/v2/perceive -H "X-API-Key: $KEY" \-H "Content-Type: application/json" \-d '{"url": "https://www.g2.com/", "outputs": ["markdown"]}'{"render_quality": 0.0, "status_code": 403, "deductions": {"empty_body": 0.7, "http_error": 0.7}, "warnings": ["crawl4ai flagged this page (Blocked by anti-bot protection: DataDome captcha); outputs were captured by our hooks and are returned anyway."], "billed": false}
curl with a Chrome user agent gets a 403 and a DataDome captcha page. /v2/perceive returns the block as a block: render_quality 0.00, the deductions that fired (empty_body, http_error), and billed: false.
Every verb, live today. Every read scored.
The perception layer between your agent and the real world: every file it must open, every page it must read, through one server and one key.
POST /v2/perceive1 op · 1 keyAny URL into clean Markdown, JSON, or a screenshot. Real Chromium rendering; lazy-load and anti-bot handled. Every render scored 0.0 to 1.0.
Your first honest read, in under a minute
You don't even have to think of a test case. We hand you the winning one.
Install
One command in Claude, Cursor or Windsurf. REST keys work everywhere else.
npx @enconvert/mcp setupPoint it at our fixture
A deliberately hostile page we host, so your first call is the impressive one.
https://fixtures.enconvert.com/hostile-spaRead the receipt
Clean markdown comes back scored. Wire one alert on scores below 0.40 and silent failure is off your list.
"render_quality": 1.02,500+ conversions and reads through the keyless playground. In the public benchmark, 0 of 50 public URLs came back as a block page presented as content. Run it yourself.
We don't solve CAPTCHAs, run residential proxies, or get around paywalls. When a page looks blocked, the render ladder retries once with a hardened browser. If it is still blocked you get a score under 0.40 and the reason, not a 200, and you are not billed for it. robots.txt is honoured when you set respect_robots.
Retire the duct tape
The real competitor isn't another API. It's the pile of vendors and glue scripts you're babysitting right now.
EnConvert — monthlyMeets your stack where it is
Agent-native first. The same verbs through whichever door fits your stack, all live today.
You're about to depend on us. Here's why that's safe.
We can't hand you a wall of logos yet. We can hand you the things logos are a proxy for: security posture, honest failure modes, and a public record of shipping.
Start free. Scale when your agent does.
One price everywhere: the numbers on this page are the numbers in the API and on the receipt.
Free while in beta. No card. 500 ops a month.
3,000 reads, 20 watchers, 45 conversion routes, $5 extraction credit.
For teams scaling an agent past the demo.
For when your agent is the product.
Running past 50,000 ops a month, or need custom retention and terms? Talk to us.
Something missing? The docs go deeper, and the founder answers email.
What exactly does render_quality measure?
A 0.0 to 1.0 score of how faithfully the page rendered before we read it. Challenge pages, empty SPA shells, partial lazy-loads and interstitials all pull it down. A low score comes back with a warnings array naming what we detected, and the full output is still returned so you decide what to trust, which means a bad read can never quietly enter your agent's context unnoticed.
What happens when a page blocks you?
You get the truth: a low score, a warnings array naming what we detected, and no garbage passed off as content. Sometimes the web wins. The product guarantee is that you will know when it does, in the same response, not from a customer three days later.
Is this production-ready?
The seven v2 verbs are live and production-hardened: SSRF screening on every URL, honest failure contracts, and a public status page and changelog you can watch before you commit. We publish what breaks and when we fix it.
How is this different from Firecrawl or CloudConvert?
They're good tools on one side of the boundary. Scraping APIs stop at pages; conversion APIs stop at files. Agents need both, so teams end up wiring two or three vendors together. EnConvert is one MCP server for both, with one bill, and it's the only layer that scores every read and refuses to pass off blocked pages as content.
Do I need a backend?
No. Public keys are domain-restricted, so they're safe to ship in client code. Secret keys exist for server-side work when you want them.
The web lies to your agent. EnConvert doesn't.
One MCP server for every file and every page, with a receipt for every read. Free while in beta. No card. 500 ops a month.
$ npx @enconvert/mcp setup