Firecrawl

/monitor shipped May 26: describe what to track in plain English and get signed webhook diffs only when pages actually change.¹

Agent-friendly page monitoring. The new /monitor endpoint configures URLs, schema, and schedule from a plain-English goal. Every change is a permalink — a first-class diff object you can share or hand off to another agent.² Delivery via signed webhook or email, with cadence options from 5-minute to custom cron.

v2.10 expanded the core API surface. /parse accepts PDFs, Word docs, and spreadsheets up to 50 MB via a Rust-based engine (up to 5x faster), returning Markdown, JSON, or a summary with tables and reading order preserved.³ Zero Data Retention is available for enterprise plans.

Two new scrape formats cut token costs sharply.

• highlights returns the exact sentences, code blocks, and table rows matching a query — nothing is rewritten or hallucinated, up to 100x fewer tokens.⁴
• question returns a grounded answer pulled directly from page content, isolated against prompt injection via XML tagging and zero-width-space escaping, also up to 100x fewer tokens.⁵

Lockdown Mode keeps sensitive requests contained. Set lockdown: true on /scrape to serve results exclusively from Firecrawl's index — no outbound requests, no stored data.⁶ Works across all SDKs, the CLI (--lockdown), and MCP.

SDK coverage reached nine official languages.⁷ Go, Ruby, PHP (.NET) joined in v2.10; Rust was promoted to the official v2 build. Java and Elixir arrived in v2.9.0. The open-source web-agent framework launched for building multi-source research agents with any model.⁸

The /interact stateful browser sessions continue to support click, form-fill, and multi-step navigation with persistent named profiles across scrapes.⁹