Pages, documents and feeds
Primary sources, any format. JavaScript pages, PDFs, Office files, Shopify or WooCommerce catalogs, and job boards in one suite.
read: Clean Markdown, highlights and inline JSON in one call.
DocumentationWeb API referenceSearch referenceInvoice reconcileOpenAPI specsllms.txt
Hosted pilotweb.vitna.ai/v1 · hosted web data suite
Read a JavaScript page. Pull out its prices. Collect a website. Watch for the next change. One API suite for the work between a URL and useful data.
Keys come by invitation while public signup is closed, activated on the account page. Nothing is billed: usage is measured monthly and payment collection is off.
POST /v1/readcurl https://web.vitna.ai/v1/read \
-H "Authorization: Bearer $VITNA_WEB_KEY" \
-H "Content-Type: application/json" \
-d '{"url":"https://example.com/","format":"markdown"}'# Example product A useful product for your daily work. Price: $49.00
Illustrativeno page was contactedPOST web.vitna.ai/v1/readnothing billed: payment collection is off
Illustrative response. No website has been contacted.
Set VITNA_WEB_KEY to your API key. This request uses the hosted endpoint at https://web.vitna.ai.
curl 'https://web.vitna.ai/v1/read' \
-X POST \
-H "Authorization: Bearer $VITNA_WEB_KEY" \
-H 'Content-Type: application/json' \
-d '{
"url": "https://example.com/",
"format": "markdown",
"render": true,
"timeoutMs": 30000,
"maxCharacters": 50000
}'Request a Web API key, or open your API account. An owner token manages the account; a Web API key, or your org’s Vitna key, executes requests.
Built for apps and coding agents
Pages, documents and feeds
Primary sources, any format. JavaScript pages, PDFs, Office files, Shopify or WooCommerce catalogs, and job boards in one suite.
read: Clean Markdown, highlights and inline JSON in one call.
MCP, REST and Markdown docs
One line into any agent. Wire Claude Code, Cursor or Codex to /v1/mcp with the same bearer key.
$ claude mcp add --transport http vitna https://web.vitna.ai/v1/mcp
--header "Authorization: Bearer $VITNA_WEB_KEY"
✓ Vitna Web API tools wired into Claude Code
Matched passages
The passage, not the whole DOM. Verbatim highlights, page-grounded questions and PII redaction come back clean in one read.
read: example.com/productPOST /v1/read · renderedThe example product is available for $49.00 with a two-year warranty and standard shipping.
onlyMainContent: trueexcludeTags: [".cookie-banner", "aside"]Search and live scrape
Find pages, then read them in one call. Query Vitna's index with domain and crawl-age filters, with optional live scraping of every hit.
scrape: true$49.00 · in_stock · live scraped in the same search call
Change monitors
Meaningful moves, not noisy diffs. Watch page selectors or search queries on an interval; suppress noise below your threshold and post real changes.
Webhook posted: price moved past meaningfulChange threshold.
Bounded budgets and filters
Scope every crawl and extraction. Bound paths, depth, sitemaps, DOM tags and PII redaction before a browser renders a page.
Keep discovery strictly inside product routes.
Illustrative previews of hosted Web API capabilities on example.com; live responses depend on the target page and your key's allowances.
Use the same bearer key over HTTP MCP, hand a coding agent the verified setup prompt, or point it straight at the Markdown specification.
Adds web.vitna.ai/v1/mcp over HTTP with your bearer key.
read · batch · screenshot · pdf · documentsextract · product · catalog · careers · map · crawlsearch · google_search · research · sourcesPoint your agent at https://web.vitna.ai/v1/mcp with your bearer key, or hand it the setup prompt written from the reference.
claude mcp add --transport http vitna https://web.vitna.ai/v1/mcp \
--header "Authorization: Bearer $VITNA_WEB_KEY"Every documentation page on vitna.ai answers as plain Markdown at its .md address, so an agent reads routes, limits and error codes without scraping HTML.
curl https://vitna.ai/docs/web/overview.mdEvery endpoint takes the same bearer key and answers in JSON: a page's Markdown, a product's fields, a crawl's pages, a screenshot's bytes. One response first, read field by field, then every endpoint by what it is for.
{ "url": "https://example.com/", "data": { "name": "Example product", "price": 49, "currency": "USD", "availability": "in_stock", "sku": "DESK-01", "brand": "Example", "images": ["https://example.com/product.png"], "offers": [{ "price": 49, "currency": "USD", "availability": "in_stock" }], "variants": [{ "name": "Example product, walnut", "sku": "DESK-02" }] }}urlThe page that was read. It stays as read even when redaction is on, because it is how a result is matched to its page.
price · currencyThe published offer as a number and a currency code. When the offers publish different prices, price is null and priceRange holds the low and high that were observed.
availabilityThe stock state the page publishes, in one word such as in_stock.
sku · brand · imagesWhat the page publishes about the product. A field it leaves out comes back null, or an empty list, and is never estimated.
offersEvery published offer, each with its own price, currency and availability.
variantsVariants stay here. A variant's price becomes the price only when the page names that variant.
This is what a key can call on the hosted pilot, Sources included. Browse indexed verticals and official data providers in Vitna Library, or read the API reference for every field, limit and error code.
/v1/read/v1/batch/v1/screenshot · /v1/pdf/v1/documents/v1/parse/v1/search/v1/extract/v1/product · /v1/company/v1/catalog/v1/careers/v1/ir/v1/entity · /v1/grid/v1/map/v1/crawl/v1/jobs/:id/v1/monitors/v1/diff/v1/recipes/draft · /v1/recipes/v1/datasets/v1/firehose · /v1/firehose/events/v1/research/v1/verify/v1/sources/find · /run/v1/mcpThree places the suite does an app's work, each drawn as the exact requests it takes.
Three requests · map, extract, monitor
A shopping app needs the product pages, a consistent price field, and a way to notice a change. That is three requests.
1 · Map
Include /products/ and leave out the pages your app does not need.
POST /v1/map
{
"url": "https://example.com/",
"includePaths": ["/products/"],
"excludePaths": ["/account/"]
}"urls": [
"https://example.com/products/desk",
"https://example.com/products/lamp"
]2 · Extract
Name a selector for the product name, the price and the stock state, and receive JSON.
POST /v1/extract
{
"url": "https://example.com/products/desk",
"selectors": {
"name": "h1",
"price": ".price",
"stock": ".availability"
}
}"data": {
"name": "Example desk",
"price": "$49.00",
"stock": "In stock"
}3 · Monitor
Watch the same selectors on an interval. The first check sets the baseline; a later check that finds a change posts it to your webhook.
POST /v1/monitors
{
"url": "https://example.com/products/desk",
"intervalSeconds": 3600,
"selectors": { "price": ".price" },
"webhookUrl": "https://your.app/hooks/price"
}"type": "monitor.changed",
"changes": [{
"field": "price",
"previous": "$49.00",
"current": "$45.00"
}]Add meaningfulChange with a minimum token-diff ratio so footer timestamps and layout tweaks never trigger false webhook alerts.
HTTP MCP · POST /v1/mcp
An agent reads pages, pulls fields, maps and crawls sites and lists a company's open roles through the tools its client already understands, with the same key. Each tool call is the REST request it stands for, counted and metered the same way.
Connect to the API host itself: the relay this page's playground uses does not forward MCP. The official data-source tools are listed with the others.
google_search returns Google's results page through Vitna Search, so it takes your org's key, vitna_key_, and counts against Search's limits.
Add to Claude Codebash
claude mcp add --transport http vitna https://web.vitna.ai/v1/mcp \
--header "Authorization: Bearer $VITNA_WEB_KEY"Or configure in JSONmcpServers
{ "mcpServers": { "vitna": { "type": "http", "url": "https://web.vitna.ai/v1/mcp", "headers": { "Authorization": "Bearer vitna_web_..." } } } }Speaks MCP protocol 2026-07-28 back to 2025-03-26 over standard HTTPS with no session handshakes or long-lived sockets to maintain.
Scheduled datasets · POST /v1/datasets
A dataset runs a saved recipe, a store's catalog or a careers board on an interval. Each run keeps its records as JSONL or CSV, with what was added, changed and removed since the run before, and a webhook hears when it completes.
Create a scheduled datasetJSON
POST /v1/datasets
{
"name": "Competitor prices",
"source": {
"recipe": "RECIPE_ID",
"urls": ["https://shop.example.com/p/1"]
},
"key": "url",
"intervalSeconds": 86400,
"webhookUrl": "https://your.app/hooks/runs"
}Every scheduled run tracks added, changed and removed rows keyed on your chosen field so downstream pipelines ingest only real deltas.
Illustrative requests and responses. Rendering, selectors and the site's own availability decide what each one can return.
Ask for an invitation, activate it in your browser, keep the two credentials it shows once, then send your first request.
https://web.vitna.ai/v1.Your first requestscURL
# Read a page as Markdown
curl https://web.vitna.ai/v1/read \
-H "Authorization: Bearer $VITNA_WEB_KEY" \
-H "Content-Type: application/json" \
-d '{"url":"https://example.com/","format":"markdown"}'
# See what today's requests used
curl https://web.vitna.ai/v1/usage \
-H "Authorization: Bearer $VITNA_WEB_KEY"The service runs on Vitna's machines at web.vitna.ai. Your side is an HTTPS request and a key; nothing of yours is installed or deployed.
01
JSON over HTTPS to https://web.vitna.ai/v1, the key as a bearer token, from your server, a script or an agent. The playground sends through vitna.ai's /api/web relay; MCP clients connect directly to /v1/mcp.
02
A vitna_web_ key is checked against the service's own key store, and an org's vitna_key_ with vitna.ai. The key's allowances admit the request, or refuse it with 429 and Retry-After.
03
Routed to one of three engines below by what the request asks for. Whichever way, destinations must be public HTTPS addresses, and every redirect and resolved address is checked again.
04
Fields come from selectors, published structured data and stated facts without a language model; research answers with cited passages. Long crawls, extractions and research runs return a job polled at /v1/jobs/:id or posted to a webhook.
POST https://web.vitna.ai/v1/*Authorization: Bearer $VITNA_WEB_KEYread · extract · product · crawl · monitors · screenshot · pdfcatalog · careersrobots.txt first, with no browser.search · discoveryJobs, monitors and datasets run in the service, so they need no process of yours to stay up: a queued job resumes after a restart, and a monitor keeps checking on its interval.
Every bound below is the reference's own, with the section that states it. The service reports the live ones for your key at GET /v1/status and GET /v1/usage.
60 s at most, 30 s unless you set it
Reader and scrape10 min at most
Crawl jobs1,000 URLs
Site map500 page attempts, 20 links deep
Crawl jobs20 pages
Structured extraction20 sources, 40 page attempts, 3 search rounds
Research16 MiB; PDF, Word, Excel or PowerPoint
Reader and scrape60 s to 30 days
Change monitorsA per-minute and a daily allowance. A request past either is refused with 429 and Retry-After.
Capabilities and usageA daily allowance, and a retry counts as an attempt.
Capabilities and usageGET /v1/usage returns them, with today's counts
Capabilities and usageThe latest 20
Crawl jobs5
DatasetsAny HTTPS client: JSON in and out, the key as a bearer token.
AuthenticationProtocol 2026-07-28 back to 2025-03-26, over HTTP, with no session
MCPPublic HTTPS pages only, every redirect and resolved address checked.
Deployment boundariesNeeds a language model. The native engine answers model_not_configured, and GET /v1/status reports promptExtraction.
Structured extractionCited passages and stated facts. A question its sources do not answer is refused, never guessed.
ResearchCleaned content, not the full browser DOM.
Reader and scrapeNot read: OCR is not available.
Reader and scrapeBrowser sessions, the news firehose, published datasets, proxy routes and Sources are switched on by the operator. GET /v1/status names the ones that are on.
Capabilities and usageRendering cannot make a site reachable, or every late field ready.
Deployment boundariesNo speed or coverage figures are printed here. The reference's benchmark runs against a local fixture, and says itself that it does not establish hosted capacity or how much of the web a read reaches.
What Vitna charges, kept apart from what anyone else would. The pricing page keeps every hosted product's price apart from the plans.
The short answers on credentials, bounds, storage and error handling. The API reference has the full request contract.
Read specification as MarkdownNot until you run one. The response it shows first is an illustrative example and says so. With a Web API key, Run sends your request to the web service configured for this website and prints what comes back.
No. Reading, selectors, published structured data and captures run in the service without one. Research on the hosted pilot is extractive: it returns cited passages from its sources rather than a generated answer. A plain-words prompt for extraction is the one thing that needs a model, and the service says so rather than ignoring it.
Within bounds. A crawl runs as a background job of up to 500 page attempts, a map returns up to 1,000 URLs, and multi-page extraction reads up to 20 pages. What comes back also depends on the site's own availability.
Nothing. Payment collection is off and no rates are set. The service measures what each request uses, browser time, render attempts, search and extraction calls, and your account shows it as a monthly usage statement, which carries no total while no rates are set.
Activating an invitation shows two, once each: an API key (vitna_web_) that runs requests, and an owner token (vitna_web_owner_) that issues, rotates and revokes keys and reads statements. An invitation lasts 24 hours by default. A key expires after 90 days unless you pick a date within 366, and an account keeps up to 20 keys. Keep both private: whoever holds a key can spend it. A lost owner token cannot be shown again; ask us and an operator issues a new one.
Yes. An org's vitna_key_, which an operator mints for the org, works at web.vitna.ai and in the playground exactly as a Web API key does. The service checks it with vitna.ai, a revoked key stops working within a minute, and the org is limited and metered as any other client. Org keys are not minted from a page yet.
It stays in the playground's memory while this page is open and goes only with the requests you run. It is never saved to browser storage. The account page holds its credentials the same way.
A result goes back to you. A job's pages are kept with the job, which belongs to your key's client and no other, and the latest 20 jobs are kept. A monitor keeps the values it watches and its last 20 checks. Set redact to take email addresses, phone numbers, card numbers, IBANs, US Social Security numbers and IPv4 addresses out of a response, and out of a crawl's pages before the job stores them. The usage ledger keeps what each request used, browser time and calls, for 400 days.
It answers with an HTTP status and a JSON error carrying a code, a message and a requestId to quote. A key past an allowance gets 429 (rate_limited, daily_limit or page_limit) with Retry-After. A crawl or a multi-page extraction that misses some pages still completes and lists them in errors, and extraction or research that cannot find a stated value refuses with a 422 rather than guessing.
Through the contact form; one person reads it and replies. If a request failed, include its requestId.
Get started · hosted pilot
Keys come by invitation while public signup is closed. A person reads each request, the invitation is activated on the account page, and nothing is billed during the pilot.
Building with a coding agent? Fetch /docs/web/overview.md, inspect /llms.txt or /llms-full.txt, or load OpenAPI 3.1 and MCP discovery.