Extract every HTML table from a page (or raw HTML) as structured JSON: headers and rows kept intact. The deterministic answer to 'get the pricing table / comparison table / stats table from this page' — no vision model needed. Use this when an agent needs to hTML tables → structured JSON rows and columns.

Split text into retrieval-friendly chunks with a token budget and sentence-aware boundaries. Overlap between consecutive chunks preserves context across splits — the standard preparation step before embedding. Use this when an agent needs to rAG-ready chunks with token budgets and overlap.

Scan unstructured text and extract structured entities: email addresses, URLs, IPv4 addresses, phone numbers and domains. Deduplicated, categorized, machine-ready. Use this when an agent needs to pull emails, URLs, IPs, phones and domains from raw text.

Estimate how many tokens a text consumes for each major model family (GPT, Claude, Gemini, Llama, DeepSeek) using a character/word heuristic that is within ~10% of real BPE tokenizers — enough for context budgeting and cost math, with zero dependencies. Use this when an agent needs to approximate token counts per LLM family, ±10%.

Percent-encode text for safe inclusion in URLs and query strings, or decode percent-encoded text back. Uses the standard encodeURIComponent rules. Use this when an agent needs to percent-encoding for query strings and URLs.

Parse a URL into all its components: protocol, hostname, port, path segments, query parameters (as a clean object), hash and credentials. Exact — no guessing by eye. Use this when an agent needs to break any URL into protocol, host, path and params.

Fetch a web page server-side and convert it to clean markdown — headings, links, lists, tables and text, with scripts, styles and boilerplate stripped. The 'reader' pattern agents use to actually understand web pages. SSRF-guarded with redirects re-validated, 10s timeout, 1MB cap. Use this when an agent needs to fetch any page and get clean markdown an LLM can read.

Parse a User-Agent header into structured data: browser (name/version), rendering engine, operating system, and device type/vendor/model. Deterministic — the same string always parses identically. Use this when an agent needs to uA string → browser, engine, OS and device.

Generate cryptographically random UUIDv4s or time-sortable UUIDv7s from a real CSPRNG, and validate any UUID's format, version and variant bits. Use this when an agent needs to random v4, sortable v7, plus strict validation.

Calculate the compound annual growth rate that takes an initial value to a final value over a given number of years — the standard way to annualize returns. Use this when an agent needs to compound annual growth rate, time-normalized.

Validate an International Bank Account Number: country code, country-specific length and the ISO 7064 MOD 97-10 check digits — with the pretty-printed grouping returned when valid. Use this when an agent needs to structure, length and mod-97 checksum validation.

Calculate gross profit, profit margin (profit as % of price) and markup (profit as % of cost) from a unit cost and selling price — the two numbers people constantly mix up. Use this when an agent needs to profit margin vs markup, both computed.

Calculate return on investment: profit (or loss) and the percentage return relative to the initial amount invested. Use this when an agent needs to return on investment as a percentage.

Count business days (Mon–Fri) between two dates, with weekend and holiday breakdown. Provide a country code and real public holidays (via Nager.Date) are excluded too. Use this when an agent needs to working days between dates, optionally minus public holidays.

Query live DNS records (A, AAAA, MX, TXT, NS, CNAME, SOA) through Cloudflare's DNS-over-HTTPS resolver. Domain validation, email deliverability pre-checks and infrastructure discovery. Use this when an agent needs to real DNS records via Cloudflare DNS-over-HTTPS.

Live foreign-exchange reference rates from the European Central Bank (via frankfurter.dev): a full rate table against any base currency, or cross-rate conversion of any amount. Refreshed every business day by the ECB, cached one hour. Use this when an agent needs to live ECB reference rates for 30 currencies.

List upcoming or full-year public holidays for any covered country via Nager.Date — names, dates and whether they're nationwide. Essential for scheduling agents. Use this when an agent needs to upcoming public holidays for ~110 countries.

Geolocate any IPv4/IPv6 address (or the caller's own IP when omitted) via ipwho.is: country, region, city, coordinates, timezone, ASN/network owner and currency. Cached 24 hours per IP. Use this when an agent needs to where is an IP address? Country, city, ASN, timezone.

Generate a QR code for any text, URL or crypto address: returns ready-to-embed SVG markup and a data URI, with configurable size and error-correction level. Use this when an agent needs to any text or URL as a QR code (SVG + PNG data URI).

Fetch a URL server-side and extract its title, meta description, Open Graph and Twitter card tags, canonical URL and favicon — the data agents need to preview or classify a link. SSRF-guarded: private networks are blocked, 10s timeout, 1MB cap. Use this when an agent needs to title, description, OG image and favicon of any page.

Live weather from Open-Meteo: current temperature, humidity, wind and conditions plus a 3-day forecast, by city name or coordinates. No API key, cached 30 minutes. Use this when an agent needs to current conditions + 3-day forecast for any city.