AI Scraper by Parsera
Developer ToolsExtract structured data from any web page with Parsera's LLM-powered scraper. Agents describe the fields they want in plain language and receive typed records back, with proxy and geolocation control for sites that resist automated access.
What This Integration Enables
Parsera is built for teams that need the same page read thousands of times rather than one page read brilliantly. That distinction shows up in the shape of the connector. There is a fast path for a single URL, where you describe the attributes you want and get typed records back with no setup at all. There is an agent path for pages that hide their content behind clicks, scrolling and multi step navigation. And then there is the part that matters at volume: an agent studies a page once, compiles what it learned into a reusable scraper, and every run after that executes the compiled scraper without a model in the loop.
Requests carry proxy country selection and session cookies, so an agent can ask for the German version of a marketplace listing or read a page that sits behind a sign in your team already holds. That reach is also why this is one of the connectors where FlowRunner's posture matters most. The agent gathers, the agent compares, and the agent stops at the point where a difference could mean two entirely different things.
Without FlowRunner
With FlowRunner
Use Case Scenarios
Supplier catalogue price and availability watch
A buying team tracks a few hundred supplier and marketplace product pages. A scheduled flow calls Run Scraper Asynchronously against a saved extractor scraper, which fans the URL list out into one child run per URL under a single run identifier. The agent polls Get Scraper Run until the status settles, reads the records keyed by the URL they came from, and diffs each price and stock field against the last stored value in PostgreSQL. Changes go to the buying channel in Slack with the source URL attached, so a buyer can open the page and see the same number the agent saw.
Regional availability check before a purchase commitment
Before committing to a supplier, an agent needs to know what the product page says in the market the goods will ship to, not what it says from a United States exit location. The agent calls Extract Data From URL with the proxy country set per market and the same attribute list each time, so the records line up across regions. Where the page needs a session, the agent supplies cookies rather than trying to reproduce a login. The result is one comparable record set per country, written into a Google Sheets tab the sourcing team already reviews.
Building a reusable scraper for a new supplier site
A new supplier is onboarded and their catalogue has a layout nobody has scraped before. The agent calls Build Agent Scraper, which sends an AI agent to study the page and compile an extraction script, then polls Get Agent Scraper until the status leaves generating. The build returns a sample of what the compiled scraper produced. The agent posts that sample next to the live page for a sourcing analyst to confirm the fields landed in the right columns. Only after that confirmation does the scraper get added to the weekly run list.
Human-in-Loop Highlight
The dangerous moment with a compiled scraper is not failure. It is silence. A compiled scraper no longer invokes a model on each run, which is exactly what makes it affordable at volume and also what removes its ability to notice that the page changed underneath it. The catalog is explicit that a scraper is tied to the layout it was built on and must be regenerated when the target site changes structure. So when a run comes back completed_partial, or when a field that has been populated for months is suddenly empty across a block of URLs, the agent does not overwrite the price table and it does not quietly retry. It posts a message to the sourcing channel that reads: "Compiled scraper for Acme catalogue returned an empty price on 40 of 312 URLs this run. Build sample from 12 March had this field populated. Is this a delisting, or did the page layout change and the scraper needs rebuilding?" It attaches the build sample from Get Agent Scraper, the three affected URLs, and the credits the run consumed. A person decides whether the number is real. The agent does not turn missing data into a price.
Agent Capabilities
19 actionsExtraction
5- Extract Data From URL Fetches a page and extracts the fields you describe as named attributes, returning one record per item found. The fastest route for a one off URL, because no scraper has to exist first. Set a proxy country when the page is geo restricted or unavailable from the United States, which is the default exit location.
- Extract Data From Content Runs the same extraction against raw HTML or plain text you already hold rather than fetching a URL. Used when the markup arrived from elsewhere in the flow, such as an HTTP request, a stored file or an email body, or when the page needs a session this connector cannot reproduce.
- Get Page Markdown Fetches a page and returns clean markdown with navigation, scripts and styling stripped out. No language model is involved, which makes this the cheapest way to pull readable page text for a summary, a vector store or a downstream prompt.
- Start Async Extraction Queues an extraction and returns a task identifier immediately instead of holding the connection open. Used for slow pages, large paginated listings, or any run that would otherwise time out.
- Get Async Extraction Result Reads the status and, once finished, the records of a queued extraction. While the work is queued or running only the status comes back, so the agent polls until the status leaves pending and running.
Agent
3- Start Agent Extraction Launches an AI agent that opens the page, navigates and interacts with it, then extracts the data described by the prompt. Chosen over Extract Data From URL for sites that hide content behind clicks, scrolling or multi step navigation, or when the page structure is unknown.
- Get Agent Extraction Result Reads the progress of an agent extraction and returns the records once it has finished. The data array is only present on a completed run, so the agent polls until the status leaves pending and running.
- Stop Agent Extraction Cancels an agent extraction that is still running so it stops consuming credits. Used as a guard when an upstream condition makes the extraction unnecessary before it finishes.
Agent Scrapers
4- Build Agent Scraper Sends an AI agent to study a page and compile a reusable extraction script that afterwards runs on similar pages without invoking the model again. The build runs in the background, so the agent polls Get Agent Scraper until the status is ready.
- Create Empty Agent Scraper Reserves a new, empty agentic scraper and returns its identifier. Used when the identifier has to be recorded or referenced before the build is triggered.
- Get Agent Scraper Returns a scraper's build status along with its name, the page it was built on, the sample output produced during the build and, where the scraper takes run time inputs, the parameters schema. The build sample is the reference an operator compares against when a later run looks wrong.
- Delete Agent Scraper Permanently removes an agentic scraper and stops any build still in progress. Used to clear out experiments or scrapers whose target site has changed beyond repair.
Extractor Scrapers
3- Create Empty Extractor Scraper Creates a new, empty extractor scraper and returns its identifier. Extractor scrapers are configured in the Parsera app and executed here, so one saved configuration covers many URLs without repeating attribute definitions in every flow.
- Get Extractor Scraper Returns the stored configuration of an extractor scraper, including its name and the primary URL it was built for. Used to confirm which saved scraper an identifier refers to before running it against a batch.
- Delete Extractor Scraper Permanently removes an extractor scraper. Any flow still referencing its identifier starts failing afterwards, so the agent confirms the scraper is unused first.
Scraper Runs
4- List Scrapers Returns every scraper on the account, both extractor and agentic, with identifier, name, type, build status, primary URL and creation time. The starting point for discovering scraper identifiers and auditing which are ready rather than generating or failed.
- Run Scraper Runs a saved scraper and waits for the records in the same call, which suits quick single URL jobs. If the work exceeds five minutes the response carries a run identifier and a running status instead. Extractor scrapers accept up to 100 URLs at once; agentic scrapers take exactly one.
- Run Scraper Asynchronously Queues a saved scraper and returns a run identifier immediately. The recommended way to execute scrapers and the required way to execute agentic ones. Extractor scrapers fan a list of up to 100 URLs into one child run per URL under a single run identifier.
- Get Scraper Run Reads progress and results of an asynchronous run. While running it reports how many queued URLs have finished; once settled the status is completed, completed_partial when some URLs failed, or failed, with records keyed by source URL and the credits consumed.
Frequently Asked Questions
What can FlowRunner do with AI Scraper by Parsera?
FlowRunner agents can run Extract Data From URL, Extract Data From Content, and Get Page Markdown in AI Scraper by Parsera, plus 16 more actions.
Does connecting AI Scraper by Parsera to FlowRunner require OAuth?
No. AI Scraper by Parsera connects to FlowRunner with an API key, no OAuth flow required.
Can AI Scraper by Parsera trigger a FlowRunner workflow automatically?
AI Scraper by Parsera doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.
Start building with AI Scraper by Parsera
$100 in credits. No card required. Connect in minutes.