Search & web

Search live software, compare listings, and prepare owned drafts. Publishing stays on the website.

Read-it-later for AI agents. Save links, bookmarks, notes, and videos as Markdown and search them.

Screenshot any URL, HTML or Markdown in a real browser, and read web pages as Markdown.

Real-estate underwriting, BRRRR, comps, valuation, distress search, rent-roll and T12 parsing.

AMSOIL catalog search, account options, and dealer-referral shopping links from A1 Synthetic Oil

PDF tools for AI agents: validate, extract, merge, split, compress, OCR, sign and redact.

78 tarot card meanings, ranked topic search and nine spreads, with source links.

Growth marketing and vibe coding: search the codices, find and join events, see the passes.

Business email on your own domain that your AI agent can read, search and draft in.

Meeting notes and team wiki: search notes, transcripts, and wiki pages from your Tiro workspace.

Fetch one public HTTP(S) URL for research, extraction, or agent workflows. Returns the final URL, upstream status, content type, and decoded content. Private, local, metadata, and reserved destinations are blocked; redirects are revalidated and response size and duration are bounded.

Book search across 40M+ editions (Open Library) by keyword, author or subject, with ISBNs and covers

Search published UK auction property listings. Omit q for latest active listings. Public source facts only; not a research report.

Resolve an exact supported Tamil divine-name spelling to source-scoped records in a pinned registry. Keeps ambiguity and name coincidence explicit. No fuzzy universal synonym search, full source text or theological certification.

+1 more from Tamil Divine-Name Disambiguation

Agent Research Trace Report performs a bounded deterministic agent research trace report operation over caller-supplied AI and search evidence.

Extract supported metrics from bounded supplied stress-strain data under a declared method. Report fit and sampling limits; no structural design approval.

+1 more from Stress-strain curve analysis

MCP registry: 138k servers crawled, handshake-validated, reliability-scored. 744 production-safe.

Emission-factor lookup: fetch the underlying factor (value, unit, methodology, source, licence) by id, activity_type, BEA code, or spend classification. Audit-grade provenance, no calculation.

Travel eSIM catalog: search plans, check availability, and get exact quotes with checkout links.

The Marketing Data Layer for AI Agents - SEO, Search Console, and Analytics

Deep version of the web page brief: crawls up to 3 same-domain pages, uses the larger model, and returns a more thorough evidence-backed report with more claims and key points. Versión profunda del informe web: rastrea hasta 3 páginas del mismo dominio, usa el modelo grande y devuelve un informe más completo, con más afirmaciones evidenciadas y puntos clave.

Extract and validate structured expense data from receipt text or a provider-neutral receipt extraction. Use when an agent needs merchant, date, tax, total, line items, confidence, duplicate detection, and arithmetic validation before creating or reconciling an expense. The result is advisory and always requires human review.

Searchable Italian winery directory and estate heritage catalog. Search by winery name, region, or terroir with wine counts and winery profiles.

One call replaces two: fetches the page, strips it to readable text, then returns exactly the fields you asked for as schema-enforced JSON (fields=price,author,published — up to 10). Add list=true to pull EVERY repeated record on the page (listings, table rows, search results) as an array — that mode costs $0.06, since it needs a much larger output budget. Built for agents that need a fact off a page, not a pile of HTML — no browser, no parser, no scraping code.

Exa's own page index, read per call — no Exa account, no subscription. Pass urls= (one, or up to 5 comma-separated) and get each page as clean text with title, author and publish date. Answers from Exa's crawl rather than our fetcher, so it reads pages a plain server-side GET is blocked from. maxChars=500..40000 (default 8000). One URL is $0.002; a batch of 2-5 is $0.008, which the 402 challenge quotes. Pairs with exa-search: search, then read the winners.

Search the live web from an agent that has no browser: pass query= and get back ranked results (title, URL, snippet, relevance score) plus a synthesized one-paragraph answer over them. maxResults=1..20 (default 5); includeAnswer=false to skip the summary and take sources only. The step before reasoning about anything that happened after the model's training cutoff — prices, releases, incidents, docs. Nothing crypto about it.

Turn ANY text into clean, schema-enforced JSON: pass text= plus the fields you want (fields=name,email,price,date — up to 10) and get exactly those keys back, guaranteed-valid JSON via structured outputs. Add list=true to extract EVERY repeated record (invoice lines, listings, table rows) as an array. Up to 16K chars per call. Not crypto-specific — the universal parse step for agent pipelines: pages, emails, receipts, logs. Many documents? ai-extract-batch: 10 in one call for $0.10.

Search current U.S. Grants.gov opportunities by default, or set source=eu to search English EU Funding & Tenders topics with SEDIA status 31094502. Inputs are source, query, agency, statuses, limit, closing_within_days, and minimum_award_ceiling. Receive up to 20 normalized JSON records with title, agency, status, open and close dates, award amounts, eligibility, funding fields, and official_url.

Exa's neural web search, resold per call — no Exa account, no subscription, one USDC micro-payment. Pass query= and get ranked results: title, URL, publish date, author, relevance score. numResults=1..10 (default 5). type=auto|fast|instant. includeDomains= restricts the search. Add text=1 for matched highlights from each page (caps results at 5, priced $0.03; the 402 challenge quotes it). For agents that need what the web says today, not what a training cutoff remembers.

Pass up to 5 URLs (comma-separated) and get each page back as clean text with its title. Renders pages that build their content in the browser, which a plain server-side fetch cannot read at all — so it covers the SPAs, dashboards and app pages that url-extract deliberately refuses. Returns partial success: pages that fail come back in a `failed` list with the reason, and a call where every URL fails is not charged. For a single static page, url-extract is cheaper.