Player not loading? Watch on YouTube
This tutorial compares Firecrawl's Scrape and Interact endpoints for web research and AI agent workflows. The speaker explains that Scrape checks for a cached URL before rendering an uncached page in headless Chrome. It executes JavaScript and returns formats such as markdown, JSON, HTML or screenshots, along with branding information.
Scrape captures the rendered page, but it also accepts scripted clicks, scrolling and key presses when you know the relevant element attributes. The speaker recommends Interact when navigation needs natural language instructions or a browser session must persist across calls. In the described workflow, a scrape ID starts an interaction with a remote Chrome instance. An LLM translates prompts into Agent Browser commands, and a CDP connection provides browser control. An iframe also lets users operate the session manually.
The examples contrast homepage extraction with adding a drink to a basket and completing checkout on a fake shop. Another example uses a persistent profile to log in, enter cell values and return an accessibility audit as JSON. The presenter uses the Node SDK and says both endpoints also work through Firecrawl's MCP server and CLI.
The cost discussion matters for session management: the speaker says Interact bills per browser minute and closes after 5 minutes of inactivity. Explicitly stopping the interaction can save credits. His recommendation is to use Scrape with browser actions for simple, repeatable tasks and Interact for more complex navigation.
