URL to Text
AITurn any web page into clean, model-ready text with the URLtoText API, returning plain text, Markdown, or structured content. Agents ground a model on a live page without parsing HTML themselves.
What This Integration Enables
This connector does exactly two things, and that is the argument for it. Convert URL To Text takes a web page and gives back plain text, Markdown or HTML with the navigation, ads and boilerplate removed. Convert PDF To Text does the same for a PDF sitting at a public URL. Everything else it offers is in service of those two outcomes. Headless JavaScript rendering handles single-page apps that return an empty shell. CSS-selector scoping isolates one region of a page. Residential and stealth proxy pools reach sites that refuse datacenter traffic, and an optional prompt can be applied to the extracted content. A YouTube URL returns the video transcript. Responses carry the page title, published date, Open Graph image and description, plus the credits the call consumed.
The value is subtraction. Every team that grounds a model on live web content ends up maintaining a parser per site, and every site redesign quietly breaks one of them without anyone noticing until the output goes strange. Removing that maintenance surface is what makes a monitoring workflow survivable. It also changes where judgment belongs: the agent handles fetching and cleaning, and the person handles the one question a fetcher cannot answer, which is whether the change it found actually means anything. That handoff is what connectors are for in FlowRunner, and it is why human-in-the-loop sits on the interpretation step rather than the retrieval step.
Without FlowRunner
With FlowRunner
Use Case Scenarios
Competitor page monitoring that does not cry wolf
A watch runs each morning across a set of competitor pricing and documentation pages. For each one the agent calls Convert URL To Text with rendering enabled and main-content extraction on, so the output is the page body without the nav bar that changes on every deploy. It diffs the result against the previous capture. Most mornings there is no difference and nothing happens. When a price or a tier name moves, the agent posts the old and new passages side by side into Slack with the capture time, and a product marketer decides whether it warrants a response.
Feeding a retrieval index without a scraping stack
A support team wants their assistant to answer from the current version of a vendor's documentation rather than a snapshot taken at onboarding. A scheduled flow walks the documentation URLs, converts each to Markdown, and writes the chunks into a Pinecone index with the source URL and published date attached to every record. Because the extraction strips chrome, the index holds prose rather than menu labels, and answers cite a page the reader can open.
Turning linked PDFs into text a workflow can act on
An inbound email links to a supplier's updated terms as a PDF. The agent passes the URL to Convert PDF To Text and gets the extracted text back, along with a warning when parts of the document could not be read, which is the honest signal that the file was a scan rather than real text. If that warning appears, the agent stops and routes the PDF to a person instead of acting on a partial reading. If the text is clean, it goes into the contract review flow with the relevant clauses pulled out for the reviewer.
Human-in-Loop Highlight
The failure mode with page monitoring is not that the fetch breaks. It is that the fetch succeeds and returns something slightly untrue. A stealth or residential proxy may be served a different page than your customers see. An A/B test can show one price to one visitor and another price to the next. Main-content extraction can trim a qualifying line that sat outside the article body and changed what the number meant. Each of those produces a clean, confident, wrong capture, which is exactly what a diff will happily act on. So the agent treats a detected change as a signal to pull the digital andon cord, not as a fact. It posts the previous text, the new text, the exact capture time, the proxy pool used and the page's own published date, and asks: "This pricing page now reads 249 where it read 199. Same URL, rendered capture, taken four minutes ago. Confirm the change is real before we brief the sales team, or re-capture without the stealth proxy?" One person spends thirty seconds looking at the live page. Nobody sends a competitive alert built on an A/B variant.
Agent Capabilities
2 actionsContent Extraction
2- Convert URL To Text Fetches a web page and converts it to clean plain text, Markdown or HTML. Supports headless JavaScript rendering for dynamic sites, AI-powered main-content extraction that strips navigation and boilerplate, CSS-selector scoping when only one region matters, residential and stealth proxy pools for pages that block datacenter traffic, and an optional prompt applied to the extracted content. A YouTube video URL returns its transcript. Returns page metadata including title, published date and Open Graph fields, plus the credits consumed. Used as the fetch step in any workflow that grounds a model on a live page.
- Convert PDF To Text Downloads a PDF from a publicly accessible URL and converts it to plain text, for files up to 10 MB. Returns the extracted text, the credits consumed, and a warning when parts of the document could not be read, which is the reliable indicator of a scanned or image-only page. Used to bring linked reports, invoices and whitepapers into a workflow, and to detect when a document needs a person rather than a parser.
Frequently Asked Questions
What can FlowRunner do with URL to Text?
FlowRunner agents can run Convert URL To Text and Convert PDF To Text in URL to Text.
Does connecting URL to Text to FlowRunner require OAuth?
No. URL to Text connects to FlowRunner with an API key, no OAuth flow required.
Can URL to Text trigger a FlowRunner workflow automatically?
URL to Text doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.
Start building with URL to Text
$100 in credits. No card required. Connect in minutes.