Firecrawl Releases Firecrawl Local, Claiming 37x Faster Scraping and 230x Less Memory
Introducing Firecrawl Local. We rebuilt the Firecrawl repo so agents can scrape locally. - 37x faster rust based scrape primitive (vs self-hosted) - 230x less memory (11 MB idle) - 100% open source Compatible out of the box with all of our MCP & SDKs.
Context
A local, open source scraper that agents can run themselves would matter if the speed and memory claims hold, because it cuts the need to send pages to a hosted API.
Firecrawl posted that it rebuilt its repo so agents can scrape locally, under the name Firecrawl Local. The company claims a 37x faster rust based scrape primitive versus self hosted, 230x less memory with 11 MB idle, 100% open source, and compatibility with its MCP and SDKs. Those figures are Firecrawl's own claims.
Deep dive
Firecrawl, the official account, said it rebuilt the Firecrawl repo so agents can scrape locally. It branded the result Firecrawl Local and listed three claims: a 37x faster rust based scrape primitive versus self hosted, 230x less memory at 11 MB idle, and 100% open source. It also said the tool is compatible out of the box with all of its MCP and SDKs, and it linked the public GitHub repo.
The speed and memory numbers are the company's. The post does not show the test setup, the self hosted baseline, or the hardware used for the 37x and 230x lines. Those should stay attributed to Firecrawl until someone else reproduces them. 'Rust based' and 'self hosted' are the company's own wording in the post.
A local scraper that agents can run without calling a hosted API is a real product direction, even if the multiples are later revised. The open source claim, if the repo matches, lets other builders inspect the code. This is a tooling launch, not a new model or a new research result.
The post includes a short video, which is not described here. There is no independent benchmark in this file.
What to expect next
Third party tests, or a methods note from Firecrawl, would confirm or shrink the 37x and 230x claims. Watch whether the GitHub repo is the full local product and whether the MCP and SDK compatibility holds for existing users. If the company later publishes pricing for a hosted companion, that should be kept separate from the open source local tool. This should stay a mid to low ranked tooling release until the performance claims are checked.
