Read the web as Markdown.

Any URL in, clean Markdown out, with an AIO score for how readable the page is to an agent. We read robots.txt first, and when a read comes back thin we say so rather than hand you an empty page.

Get a free key Try it without one

Quick start[01]

One call, and what comes back.

1,000 free credits a month; a read is one of them. Failed reads are refunded.

requestPOST

$ curl https://api.buildonto.dev/v1/read \
    -H "Authorization: Bearer onto_sk_live_…" \
    -H "Content-Type: application/json" \
    -d '{"url": "https://stripe.com/pricing"}'

response200

{
  "status": "success",
  "url": "https://stripe.com/pricing",
  "markdown": "# Pricing & Fees\n\n> Find Stripe fees…",
  "metadata": { "title": "Pricing & Fees", … },
  "stats": {
    "raw_html_size_kb": 698.8,
    "markdown_size_kb": 16.9,
    "reduction_percent": 97.6, …
  },
  "cache": { "hit": false, "ttl_seconds": 3600 }
}

One real read of stripe.com/pricing, 24 Sep 2026. Its AIO score, 90, comes from /v1/score or /v1/read-and-score.

No key required[02]

The reader is open.

Put any URL after the host. Agents get plain Markdown; a browser gets a readable page. Same URL either way.

md.buildonto.dev/any-url, no sign-up. It's the same reader as /v1/read, with the same robots.txt check.

md.buildonto.dev/stripe.com/pricing open →

30 a minuteper IP address, with no total cap and no sign-up
1 houra page is cached, so a repeat doesn't fetch it again
No JSit reads the HTML the server sends; client-rendered pages come back flagged

Endpoints[03]

What you can call.

JSON in, JSON out, with Authorization: Bearer and your key. Scores never cost a credit.

POST/v1/readAny URL as clean Markdown1 credit
POST/v1/scoreThe AIO score: how readable a page is to an agentFree
POST/v1/read-and-scoreBoth, in one round trip1 credit
POST/v1/batchUp to 50 URLs, or a whole site, in one call1 per URL
POST/v1/mapA site's URLs, from its sitemap or links1 credit
POST/v1/extractThe structured data a page declares (JSON-LD, OpenGraph, meta)1 credit
GET/v1/usageYour quota and what you have usedFree
POST/mcpThe hosted MCP server, for Claude, ChatGPT and any MCP clientAs the tool

Errors[04]

When it fails, you get a name.

Every error comes back as { "status": "error", "code", "message" } with the HTTP status below.

Your requestFix the call, the key or the plan.

400INVALID_REQUESTThe body is missing a field or malformed

400INVALID_URLNot a parseable http(s) URL, or a private address

401UNAUTHORIZEDMissing or invalid API key

402PAYMENT_REQUIREDThe plan's credits are used and there are no pack credits left

403PLAN_RESTRICTEDNot on your plan: credit packs need a paid plan

429RATE_LIMITEDThe free tier's monthly credits are used

429CONCURRENT_LIMITToo many requests in flight at once for your plan

The siteThe target said no, or isn't a page. Failed reads are refunded.

403ROBOTS_BLOCKEDIts robots.txt disallows GPTBot or every crawler, and we honour that

403WAF_BLOCKEDIt refused our fetcher (401 or 403)

404URL_NOT_FOUNDThe page 404s, or the host doesn't resolve

404NO_RESULTSMap or batch found no pages on it

415UNSUPPORTED_TYPENot a document: an image, video or archive

413TOO_LARGEThe body is over the read limit

422IMAGE_PDFA scanned PDF with no text to extract

502TLS_ERRORIts certificate is invalid

504TIMEOUTIt took more than 15 seconds to answer

UsTell us: founder@buildonto.dev

500EXTRACTION_FAILEDThe page loaded but we couldn't convert it