Web Fetch — Any URL to Markdown or Text for AI Agents

Fetch any URL as clean Markdown and plain text for AI agents, MCP and RAG: web pages, PDF, Word, Excel, PowerPoint, EPUB, CSV and JSON. One url in, one result with title and token estimate out. Static HTML, no JavaScript rendering. $1 per 1,000 pages.

Fetch any URL and get it back as clean Markdown and plain text: one URL in, one result out. It is a web fetch tool for AI agents, MCP clients (Claude, Cursor, ChatGPT), LangChain/LlamaIndex RAG pipelines and n8n/Make/Zapier flows. Give it a url (or a list) and it works out what the link really is, then returns LLM-ready Markdown and plain text in the same row, with the title, metadata, a tokenEstimate and the route used. Web pages keep only the main content (Mozilla Readability), with headings, lists, tables and links. Documents work in the same call: PDFs (text layer, up to 100 pages by default), Word, Excel and PowerPoint, OpenDocument, EPUB, CSV and JSON. OCR for images and scanned PDFs is one toggle away, off by default so calls stay fast and cheap. What it does not do: it fetches static HTML with no browser, so it does not render JavaScript and does not get around bot protection, CAPTCHAs or logins; app-shell pages and refused requests come back as free error rows. It sends an honest crawler user agent and respects robots.txt. $1 per 1,000 web pages, $2 per 1,000 documents ($3 per 1,000 OCR pages when you turn OCR on), with no compute bill on top. Failed fetches, unsupported files, pages blocked by robots.txt and empty results are never charged.

What data does it return?

Every run outputs clean, schema-validated rows like these:

url	detectedFormat	route	title	tokenEstimate	markdown
https://arxiv.org/pdf/1706.03762	pdf	pdf-text	Attention Is All You Need	9600	Provided proper attribution is provided, Google hereby grants permission to reproduce the tables and figures…
https://en.wikipedia.org/wiki/Retrieval-augmented_generation	html	html	Retrieval-augmented generation - Wikipedia	5811	Retrieval-augmented generation (RAG) is a technique that enables large language models…
https://calibre-ebook.com/downloads/demos/demo.docx	docx	office	DOCX Demo	2604	# Demonstration of DOCX support in calibre\n\nThis document demonstrates the ability of the calibre DOCX Input plugin…
https://raw.githubusercontent.com/apache/poi/trunk/test-data/spreadsheet/SampleSS.xlsx	xlsx	office	Sample Spreadsheet	96	## First Sheet\n\n| Test spreadsheet | column_B |\n| --- | --- |\n| 2nd row | 2nd row 2nd column |…

How much does it cost?

Transparent pay-per-event pricing — you only pay for what you get, and the actor always respects your run budget: set a maximum total charge and it stops cleanly at the cap instead of overrunning.

Event	Price	Charged	Worked example
page-result	$0.001	per web page converted to Markdown	1,000 web pages = $1.00
document-converted	$0.002	per PDF (each started 50 text pages), Word, Excel, PowerPoint, OpenDocument, EPUB, CSV or JSON file converted	A 15-page PDF = $0.002; a 120-page PDF = $0.006; 1,000 Word files = $2.00
page-ocr	$0.003	per image, or scanned PDF page, read by OCR	A 10-page scanned PDF = $0.03
apify-actor-start	$0.00005	per run per GB of memory (platform start fee)	One run at 1 GB = $0.00005

Input options

Field	Type	Required	Default	Description
url	string	no	—	One link to fetch: a web page, PDF, Word (.docx), Excel (.xlsx), PowerPoint (.pptx), OpenDocument, EPUB, CSV, JSON or image. The file type is detected automatically.
urls	array	no	—	Optional: more links to convert in the same run (up to 1,000 in total, duplicates removed).
includeText	boolean	no	true	Add a text field with the Markdown syntax removed (on by default, so each row has both).
onlyMainContent	boolean	no	true	Keep only the article or main content of web pages (navigation, footers and cookie banners removed). Turn off to convert the whole page body.
ocr	boolean	no	false	Read text from images (PNG, JPEG, TIFF, WebP, GIF, BMP) and from PDF pages that have no text layer, with Tesseract OCR ($0.003 per page). Off by default here so agent calls stay fast and the price predictable; images and scanned PDFs are then reported, not charged.
ocrLanguages	array	no	["eng"]	Languages in your scanned documents (up to 3). For 29 languages and word boxes use the OCR to Text actor.
maxOcrPages	integer	no	20	Upper limit of scanned pages read by OCR in one PDF ($0.003 each). Pages beyond it are left out and the row is marked truncated.
maxPdfPages	integer	no	100	Read at most this many pages of each PDF.
maxRowsPerSheet	integer	no	1000	Excel, OpenDocument and CSV tables are cut after this many rows (marked truncated).
respectRobotsTxt	boolean	no	true	Skip URLs that the site's robots.txt disallows. It is checked before every request: web pages, direct file links and every redirect target. A robots.txt that answers with a server error counts as disallow-all (RFC 9309).

Limits & honest notes

• Web pages are fetched as static HTML, with no browser. Pages that build their content with JavaScript (an empty app shell) are reported, not charged; use a browser-based crawler for those. This actor converts the URLs you give it and does not crawl sites (see Website to Markdown for that).
• OCR is off by default in this listing. With it off, images and PDF pages without a text layer are reported with a clear reason and not charged. Turn on "ocr" to read them (up to 20 scanned pages per PDF by default, set by maxOcrPages; 7 languages). For 29 languages and word boxes, use OCR to Text.
• Legacy binary Office files (.doc, .xls, .ppt), RTF, OpenDocument presentations (.odp), HEIC/AVIF images and SVG are not supported. They are reported with a clear reason and never charged.
• Spreadsheets are cut at 1,000 rows per sheet by default (maxRowsPerSheet), and very large results are cut at about 3.5 million characters. Cut rows are marked truncated: true.
• PDFs are read up to their first 100 pages by default in this listing; raise maxPdfPages (up to 2,000) for longer PDFs.
• Files up to 50 MB. Password-protected PDFs and pages behind a login can't be read.
• Polite by design: it sends an honest crawler user agent (ActorForge-AnyUrlToMarkdown) and does not pose as a browser. When a site answers 429 (rate limited) or 403 (refused), that URL is retried at most once (429 only, after Retry-After up to 30 s) and the remaining URLs on the same host are skipped as free error rows.
• tokenEstimate is characters divided by 4. That is a good guide for English with GPT and Claude tokenizers, not an exact count.
• Respects your run budget: set a maximum total charge and processing stops cleanly at the cap. No result is delivered without being paid for, and none is charged twice.

Related ActorForge actors

• Any URL to Markdown — the same converter with OCR on by default
• Website to Markdown — crawl a whole site, not single URLs
• PDF to Markdown — PDF-only, with per-page output and RAG chunks

Changelog

• 2026-10-22 — Initial release of the Web Fetch listing: the Any URL to Markdown engine with Markdown + text in one row, OCR off and a 100-page PDF cap by default.

---

Built and maintained by ActorForge. Reliability first: every actor in the portfolio is monitored by an automated nightly self-test, and breakages are typically fixed the same day.