Context.dev API

Scrape anything from a URL on the internet. Returns the outputs you enable in formats. Handles PDFs, DOCX, PPT, XLSX, and 40 other file formats.

Find news about a company by name, domain, ticker, or ISIN, or search all news with a natural-language query. Filter articles and continue through results with a cursor.

Research the web and return a sourced answer in your JSON shape. Choose `fast` for a short task or `ultra` for deeper research.

Convert uploaded file bytes into Markdown and optional HTML.

Downloads a resource and returns its bytes as base64. Without waitForMs, returns the original HTTP response without image conversion, text extraction, or character-encoding changes. HTTP compression is decoded before base64 encoding. Supply waitForMs to render HTML with JavaScript in the browser and return the resulting HTML as UTF-8 bytes after the wait. Non-HTML resources, including images and PDFs, keep their original bytes and do not incur a browser wait. Follows public redirects and retries failed downloads through ISP and residential proxies, with a direct fallback. When country is specified, only a residential proxy in that country is used. Supply headers such as Referer for images that require a referring page. Cached results are reused according to maxAgeMs (default: 3 days; maximum: 1 year). Set maxAgeMs=0 to fetch fresh and refresh the cache. Cache identity includes the exact URL, country, waitForMs, and normalized outbound headers. Credential-bearing headers and zero data retention bypass cache reads and writes. cache_metadata reports hit, miss, or zdr and the cached result age in milliseconds. Maximum decoded resource size: 50 MiB (52428800 bytes), before base64 encoding. Successful requests cost 1 credit; errors are not billed.

Scrapes the given URL and returns the HTML content of the page. Optional extractRules return deterministic structured data in extracted using CSS selectors, attributes, lists, and nested rules, without an LLM or additional credits. Rules run on the returned HTML after selector and main-content filtering. Send extractRules as a JSON-encoded query parameter. The base request costs 1 credit; requests with browser actions cost 2 credits. A request that hits its timeoutOpts.milliseconds deadline fails with 408 and is not billed, unless timeoutOpts.behavior=return-partial is set — then the page as rendered so far is returned with `finalDOMState: "still-loading"` and billed at the base cost of 1 credit.

Scrapes the given URL into LLM usable Markdown. Inspect key_metadata on JSON responses from a recognized API key; use error_code to distinguish stable failure categories. ### YouTube YouTube URLs return the video or channel itself rather than the surrounding player and navigation chrome. A URL addressing a single video (`/watch`, `youtu.be`, `/shorts`, `/embed`, `/live`) returns its title, channel, duration, view count, keywords, full description, and the transcript when the video has captions that can be retrieved; videos without captions return everything except the transcript. A channel URL (`/channel/UC…`, `/@handle`, `/c/…`, `/user/…`) returns its name, handle, subscriber count, video count, and full description. When `includeImages=true`, video responses also include the thumbnail and channel responses include the avatar. Costs the same as any other scrape. ### Billing & errors | HTTP status | Billed? | Meaning | | --- | --- | --- | | 200 | Yes — 1 credit, or 2 credits with actions | Successful scrape, including a zero-length result when includeSelectors matched nothing. A partial result (`finalDOMState: "still-loading"`, only with timeoutOpts.behavior=return-partial) is billed at the base 1 credit with no OCR or actions surcharge | | 400 | No | Invalid input, skipped PDF, or the page could not be scraped. error_code WEBSITE_BLOCKED specifically means the site answered with an anti-bot challenge, CAPTCHA wall, or login shell instead of the page (even when the site returned HTTP 200) — retrying later or from another country sometimes succeeds. error_code NO_PUBLISHED_MARKDOWN means sourcePreference excludes html-conversion and the page publishes no Markdown version | | 401 / 403 | No | Invalid/disabled key, insufficient permissions, or credits exhausted; inspect error_code | | 404 | Yes — 1 credit, or 2 credits with actions | Target page returned or fingerprinted as not found | | 408 | No | Request timed out. With timeoutOpts.behavior=return-partial this only happens when nothing usable had rendered by the deadline | | 413 | No | Target content exceeds the maximum supported size (50 MB) | | 415 | No | Unsupported content type | | 429 | No | Concurrency or per-minute rate limit exceeded; honor Retry-After | | 500 | No | Internal error |

Extract image assets from a web page, including standard URLs, inline SVGs, data URIs, responsive image sources, metadata, CSS backgrounds, video posters, and embeds. The base request costs 1 credit, or 2 credits with browser actions. When enrichment is enabled, the entire call costs 5 credits, including requests that also use actions.

Crawl an entire website's sitemap and return all discovered page URLs. Set `includeSubdomains=true` to also discover public pages and sitemaps on child hosts such as `docs.example.com` or `brand.example.com`. Pass `search` to have the discovered URLs filtered down to the pages about a phrase (for example `pricing and plans` or `api authentication docs`), most relevant first — a searched crawl scans the whole sitemap and costs 2 credits instead of 1.

Discover a site's URLs, with page titles, descriptions, keywords, and language when available. Metadata can be missing on newly discovered URLs.

Crawl a website and return page content as Markdown. Use a batch for crawls beyond 500 pages.

Extract website data using your JSON Schema and instructions.

Search the web and optionally return page content or relevant passages with each result.

Analyze a company's landing page and web search evidence to return direct competitors for the same product or market.

Find a person from identity clues and return their profile with a match score. Available on all plans; free or disposable email addresses return 422.

Retrieve and normalize a person profile from identifiers. Education and experience entries include LinkedIn-derived organization data (website, logo, employee counts, followers, headquarters, and Context.dev industry classification) when available.

Retrieve logos, backdrops, colors, industry, description, and more from any domain

Retrieve logos, colors, company details, and social links using one lookup identifier. A direct URL limits extraction to that page.

Find up to 10 brands by name or domain, ordered by popularity. Optionally filter by country and primary language; both filters must match when provided together. Results include description, country, and language when available. Use the returned domain to retrieve a full brand profile.

Returns a simplified version of brand data containing only essential information: domain, title, colors, logos, and backdrops. Optimized for faster responses and reduced data transfer.

Extract colors, typography, spacing, and component styles from a website.

Identify fonts and available font files used by a website.

Capture the given HTTP or HTTPS URL with configurable viewport, full-page capture, wait time, popup handling, theme, scroll offset, cache age, country, and request timeout. Defaults to a 1920x1080 viewport, a 3-second wait, and a cache age of 3 days. With timeoutOpts.behavior=return-partial, a screenshot of the page rendered so far may be returned; inspect finalDOMState to identify an incomplete render. Successful requests cost 1 credit; errors are not billed.

Capture a screenshot of a website.

Classify a company into NAICS industry codes.

Classify a company into SIC industry codes.

Given a single URL, determines if it is a product page and extracts the product information.

Extract product information from a brand's website. We will analyze the website and return a list of products with details such as name, description, image, pricing, features, and more.

Queue brand or styleguide data so a later lookup can return sooner.

Scrape up to 100,000 URLs, or crawl a site, asynchronously using the same formats and scrape parameters as POST /web/scrape. Choose Markdown, HTML, screenshots, images, bytes, parsed fields, products, highlights, and JSON. Poll the batch ID or receive a webhook when it finishes.