Firecrawl Local: A Rust Scraper That Keeps Crawling on Your Machine
A web-scraping workflow can quietly become an infrastructure project. A crawler needs a queue, a browser worker, storage for crawl state, retries, HTML cleanup, and somewhere to expose the results. That is reasonable for a production service, but excessive when a developer only wants to collect documentation or give a coding agent a reliable way to read a website.
Firecrawl Local takes a different route: it packages local scraping, crawling, and site mapping into a Rust binary. Firecrawl announced it on November 18, 2026. In the November 20 GitHub Trending snapshot at 06:58 UTC, the main Firecrawl repository appeared at daily position #6 with 825 new stars recorded by the archive. A Reddit launch discussion also attracted active interest. This is a current momentum signal; Firecrawl itself is an established project, while Local is the new capability.
Firecrawl organization avatar, used only to identify the project. Source: Firecrawl on GitHub.
What Firecrawl Local actually runs
Local is not a smaller Docker Compose deployment of the hosted platform. It is a single executable that performs the core fetch, HTML-cleaning, link-extraction, and crawl operations on the machine where it runs. It can be used in three ways:
- As a CLI for one-off scrapes, crawls, and maps.
- As a local HTTP service exposing selected Firecrawl v2 API routes.
- As an MCP server that a compatible coding agent can call.
The project reuses Rust-native parsing and cleaning code from the wider Firecrawl repository. That matters because a local page and a cloud page can pass through the same underlying cleanup implementation instead of two unrelated parsers.
The mental model is:
Developer / MCP client
|
v
firecrawl-local
|-- scrape / crawl / map
|-- HTML cleanup and link extraction
|-- optional local browser for JavaScript pages
|
+--> local result on localhost
|
+--> optional Firecrawl Cloud hand-off
The diagram is conceptual, not a screenshot of the product.
Install and make your first request
You need Rust and a C compiler. Chrome or Chromium is optional; it is useful for pages that require JavaScript rendering. Node.js 22 or newer is only needed when you use the MCP launcher.
The published crates.io version checked for this run was 0.1.0. The repository's development branch can move ahead of the published crate, so pin the registry version when you want a reproducible install:
cargo install firecrawl-local --version 0.1.0
firecrawl-local --version
Try a simple page before connecting an agent:
FIRECRAWL_CLOUD=off firecrawl-local scrape https://example.com --markdown
The --markdown flag prints the extracted content as Markdown. Without it, the command returns the structured response. For a site crawl, the CLI also provides a crawl command; for discovering URLs without fetching every page, use map.
To expose the local API:
FIRECRAWL_CLOUD=off firecrawl-local serve
The service listens on loopback port 3002. Check that it is responding:
curl http://127.0.0.1:3002/health
The local API implements the scrape, map, crawl, and batch-scrape routes from the v2 interface. Existing Firecrawl SDKs can be pointed at the local base URL, but do not assume every hosted endpoint exists locally.
The privacy boundary: local does not mean every feature is offline
This is the most important detail to understand before using it with private material.
The default cloud-routing mode can hand off work that the local process cannot complete, including web search, Alexandria data-provider lookups, and pages that block a direct local fetch. Setting FIRECRAWL_CLOUD=off disables that hand-off. That is the right starting point when you need a strict local-only experiment, but cloud-dependent routes will then be unavailable.
A practical policy is to decide per workload:
- Public documentation: local-only mode is often enough.
- Pages that require browser rendering: install a supported local browser and test the target site.
- Search or blocked pages: decide whether sending the URL or query to Firecrawl Cloud is acceptable before enabling fallback.
- Private internal sites: keep cloud routing off unless your security review explicitly approves it.
The README also describes browser isolation and connection checks intended to reduce unsafe navigation. Treat those as design features to review, not as a substitute for your own security assessment. Keep the service bound to loopback unless you have added authentication and network controls.
Where the smaller design has limits
A single process is convenient, but it changes the operational model. Crawl state is held in memory rather than in a durable queue, and jobs are not a replacement for a long-running distributed crawler. Local mode also does not provide full parity with the hosted product: search and some enrichment paths depend on cloud services, while unsupported formats should not be assumed to work merely because a hosted SDK exposes them.
Choose the tool that matches the job:
| Need | Better fit |
|---|---|
| Scrape a few public pages from a laptop or agent | Firecrawl Local |
| Search, enrichment, or blocked-page handling | Firecrawl Cloud, with an explicit data policy |
| Durable, high-volume crawling with workers and queues | A production crawler stack or the full self-hosted deployment |
| One site with custom extraction rules | Playwright plus a focused parser such as Readability or Trafilatura |
The repository is licensed under AGPL-3.0. If you modify it and offer the modified service over a network, review the licence obligations before embedding it in a commercial product.
Who should try it?
Local is a good fit for developers building documentation assistants, research agents, local RAG experiments, and scripts that need clean Markdown without operating a crawler cluster. It is also useful for testing an MCP workflow before deciding whether a hosted service is necessary.
It is not automatically the best choice for a production crawler that needs durable scheduling, distributed concurrency, audit trails, or guaranteed access to difficult websites. Test representative pages, measure memory and latency on your own workload, and decide whether cloud fallback is acceptable.
Firecrawl's launch announcement advertises large speed and memory improvements over its older self-hosted stack. Those are vendor-reported comparisons; the launch material does not provide enough benchmark detail to treat the multipliers as universal guarantees. Measure pages successfully extracted per minute, failure rate, and memory under the same workload before making an architecture decision.
Sources and further reading
- Firecrawl repository and Local source
- Introducing Firecrawl Local, November 18, 2026
- Published Rust crate documentation
- Firecrawl Cloud documentation
- Community launch discussion
No verified exact-project YouTube walkthrough or X post was available during this run, so neither is embedded.
Your turn: would you trust a local scraper with internal documentation, or would you require a separate network boundary even when cloud routing is disabled?
Prepared with AI assistance and checked against the linked primary sources.
Top comments (0)