OpenGraph.io
Developer ToolsTurn any public URL into structured data with OpenGraph.io, unfurling link metadata, scraping raw HTML, and extracting selected elements. Agents preview and validate links before they are published.
What This Integration Enables
It is worth being plain about what this connector is not. OpenGraph.io does not crawl a site, does not follow links, and does not run a job you poll. It takes one URL and returns one shaped result, synchronously. That narrowness is the reason to use it. When the requirement is to understand a single link, rendering it as a preview card, checking where it actually resolves to, pulling its clean text for a language model, or capturing what it looks like, a crawler is the wrong instrument and a browser agent is an expensive one.
Within that scope the connector is genuinely capable. Unfurl URL Metadata merges OpenGraph tags, Twitter Card data and HTML-inferred metadata into a single object ready to render. Convert Page To Markdown and Extract Page Content produce clean text for ingestion, with the option of a prompt-injection safety pass over the result. Query Page With AI answers a natural language question about a page and can return strictly shaped JSON. Every operation supports cache control, headless JavaScript rendering for single page apps, and proxy escalation through datacenter, residential and mobile tiers for pages behind bot protection. Screenshots come back on a temporary URL that expires after 24 hours, so the save option copies the image into FlowRunner file storage and hands back a link that outlives the request.
Without FlowRunner
With FlowRunner
Use Case Scenarios
Validating every outbound link before a post goes live
A content flow takes a queued post and unfurls each outbound link. The agent reads the merged preview object to confirm a title, description and image exist, and reads the request diagnostics from Scrape Page HTML to confirm the redirect count and resolved host match what the writer intended. Links that resolve somewhere unexpected, or return a response code that is not a success, are held rather than published. The findings are posted into Slack as a checklist against the draft in WordPress.
Turning a shared link into a usable record
Someone drops a link into a channel and it needs to become something structured. The agent calls Query Page With AI with a response structure, asking a specific question such as which pricing tiers the page lists and what each includes, and receives shaped JSON rather than prose. That JSON is written into Notion as a competitor record with the source URL and the capture date attached, so the next person to read it knows how old it is.
Feeding clean page text into a retrieval pipeline
An assistant needs source material from a set of public articles. Rather than storing raw HTML, the agent calls Convert Page To Markdown, which strips navigation, ads and boilerplate and returns either the raw Markdown or a structured envelope carrying the Markdown plus page metadata, character and token usage and fetch diagnostics. Output length is capped so a single unusually long page cannot dominate the index, and the optional prompt-injection safety pass runs over content that came from a source nobody controls.
Human-in-Loop Highlight
The gate here is small and specific, which suits a connector that is small and specific. Unfurl URL Metadata offers AI enhancement for pages that expose poor or missing metadata. That is a genuinely useful fallback and also the one place this connector writes something the source page never said. A card built that way carries a title and description composed by a model, attributed by every reader to the publisher whose logo sits next to it. Publishing that unreviewed to a customer facing channel is not a formatting risk, it is putting words in somebody else's mouth.
So the agent separates the two cases instead of treating a filled field as a solved problem. Cards built from a page's own OpenGraph tags publish without ceremony. Cards where enhancement supplied the wording become an automation exception and stop: "The link to acme.com/whitepaper has no OpenGraph title or description of its own. AI enhancement generated: 'The Complete Guide to Warehouse Automation'. The page heading reads 'Warehouse Ops in 2026'. Publish the generated wording, replace it, or drop the preview and post the bare link?" An editor decides in seconds and the post goes out. That is the whole shape of human-in-the-loop on a connector like this one: not a gate on every action, a gate on the one action that invents.
Agent Capabilities
7 actionsMetadata
2- Unfurl URL Metadata Retrieves OpenGraph tags, Twitter Card data and HTML-inferred metadata such as title, description, image, favicon and site name for any public URL, returned alongside a merged object ready to render as a link preview. Supports JavaScript rendering for single page apps, cache control, proxy escalation, and optional AI enhancement when a page exposes poor or missing metadata.
- Get Embed Card Builds an embeddable oEmbed representation of any URL and returns a ready-to-use HTML snippet. Native providers such as YouTube and Vimeo are resolved first, then oEmbed discovery, and finally a hosted fallback card generated from the page metadata. Layout, theme and maximum dimensions of the fallback card can be controlled.
Scraping
4- Scrape Page HTML Fetches the complete raw HTML of any public page along with request diagnostics: redirect count, resolved host, response code and content type. Used when metadata alone is not enough and the full document is needed for custom parsing. Optional headless rendering executes JavaScript first, and proxy tiers help with pages behind bot protection.
- Extract Page Content Pulls clean, model-ready content out of a page: a concatenated plain text string built from the requested HTML tags plus, optionally, structured values captured with your own CSS selectors. Supports waiting for a selector, scrolling to load lazy content and clicking a load more button before capture, and can run a prompt-injection safety pass over the result.
- Query Page With AI Asks a natural language question about any public page and returns an answer grounded in that page's content. An optional response structure returns strictly shaped JSON instead of prose, which makes the result easy to map into downstream steps. A larger model size suits complex pages or multi-part questions.
- Convert Page To Markdown Converts a page into clean Markdown with navigation, ads and boilerplate stripped out, suited to model ingestion and content pipelines. Returns either the raw Markdown or a structured envelope carrying the Markdown plus page metadata, character and token usage and fetch diagnostics. Output length can be capped and a prompt-injection safety pass applied.
Screenshots
1- Capture Website Screenshot Captures a screenshot of any public page as JPEG, PNG or WebP at a chosen device viewport, optionally full page, cookie banner free, in dark mode, or narrowed to a single CSS selector. The API returns a temporary image URL that expires after 24 hours, so the save option copies the image into FlowRunner file storage and returns a durable link.
Frequently Asked Questions
What can FlowRunner do with OpenGraph.io?
FlowRunner agents can run Unfurl URL Metadata, Scrape Page HTML, and Extract Page Content in OpenGraph.io, plus 4 more actions.
Does connecting OpenGraph.io to FlowRunner require OAuth?
No. OpenGraph.io connects to FlowRunner with an API key, no OAuth flow required.
Can OpenGraph.io trigger a FlowRunner workflow automatically?
OpenGraph.io doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.
Start building with OpenGraph.io
$100 in credits. No card required. Connect in minutes.