By the fastCRW team · Tutorial · Every command below was run on 2026-11-20 against fastCRW 0.46.3 in a fresh home directory.
Firecrawl Local, launched on November 18, 2026, is one Rust binary that serves the Firecrawl v2 API on localhost:3002. fastCRW serves the same v2 routes from one Rust binary too. That means moving between them is a server swap, not a code change: start fastCRW on port 3002 and your Firecrawl SDK calls keep working. What changes is where the hard requests go. Firecrawl Local sends search and pages that block your machine to Firecrawl Cloud; fastCRW keeps them on infrastructure you run. For the full side-by-side, see Firecrawl Local vs fastCRW.
Step 1: install fastCRW
The installer downloads a prebuilt binary for macOS or Linux, on Intel or ARM. There is no Rust toolchain or C compiler to install first.
curl -fsSL https://fastcrw.com/install | sh
crw --version
Our run printed crw 0.46.3. A first scrape works straight away:
crw https://example.com
Pages that need JavaScript need a local renderer. Run crw setup and pick the local option to add one, or use the Docker stack described in the self-hosting guide, which bundles a renderer and a search sidecar.
Step 2: start the API on the port you already use
Firecrawl Local's serve listens on port 3002. Start fastCRW on the same port and bind it to loopback:
crw serve --host 127.0.0.1 --port 3002
fastCRW binds to 0.0.0.0 by default and logs a warning when no API keys are set, so keep --host 127.0.0.1 for a local-only server, or set [auth] api_keys before you expose it. The health check answers like this:
curl -s http://localhost:3002/health
{"active_crawl_jobs":0,"status":"ok","version":"0.46.3"}
Step 3: run your existing code unchanged
This is the Python example from Firecrawl Local's README, with nothing edited. We ran it against fastCRW with firecrawl-py 4.57.0:
from firecrawl import Firecrawl
app = Firecrawl(api_url="http://localhost:3002")
print(app.scrape("https://example.com").markdown)
Output:
# Example Domain
This domain is for use in documentation examples without needing permission. This is not a service; avoid relying on it for testing and monitoring purposes.
The same client also ran map, crawl and batch_scrape against fastCRW; all three completed. The JavaScript example from the same README, run with the firecrawl npm package 4.54.0, printed the same markdown:
import { Firecrawl } from "firecrawl";
const app = new Firecrawl({ apiUrl: "http://localhost:3002" });
const doc = await app.scrape("https://example.com");
console.log(doc.markdown);
Plain HTTP works the same way:
curl -s http://localhost:3002/v2/scrape \
-H 'content-type: application/json' \
-d '{"url":"https://example.com","formats":["markdown"]}'
The response is the Firecrawl envelope, {"success":true,"data":{"markdown":"# Example Domain ...","metadata":{...}}}, with statusCode, sourceURL and creditsUsed in the metadata.
The one line, if you keep the default port
fastCRW's default port is 3000. If you start it with a plain crw serve, the only edit in your code is the API URL:
- app = Firecrawl(api_url="http://localhost:3002")
+ app = Firecrawl(api_url="http://localhost:3000")
We ran both the Python and JavaScript clients on port 3000 as well; both returned the page.
Crawls and maps
A crawl returns a job ID that you poll, exactly as on Firecrawl:
curl -s http://localhost:3002/v2/crawl \
-H 'content-type: application/json' \
-d '{"url":"https://example.com","limit":5}'
{"success":true,"id":"116c19e4-...","url":"http://localhost:3002/v2/crawl/116c19e4-..."}
curl -s http://localhost:3002/v2/crawl/116c19e4-...
{"success":true,"status":"completed","total":1,"completed":1,...,"data":[{"markdown":"# Example Domain ..."}]}
POST /v2/map returns {"success":true,"links":[{"url":"https://example.com"}]} for the same site.
Step 4: swap the MCP server
Firecrawl Local's mcp command starts its local API and runs the firecrawl-mcp npm package through npx, so it needs Node.js 22 or newer. fastCRW's MCP server is built into the binary. Replace the old entry in your client's config with this, using the full path that which crw prints:
{
"mcpServers": {
"crw": {
"command": "/usr/local/bin/crw",
"args": ["mcp"]
}
}
}
We started crw mcp with no Node.js on PATH, listed its tools and called crw_scrape on https://example.com; it returned the page as markdown. crw setup can also write this entry for Claude Code, Cursor, Codex, Gemini CLI, OpenCode and Windsurf when they are installed. More detail is in Firecrawl MCP without Node.js.
What changes after the move
- Search stays yours. Firecrawl Local forwards
/v2/searchto Firecrawl Cloud. On fastCRW, the plain binary answerssearch_disableduntil you point it at a search backend you run; the Docker stack starts one for you. A fastCRW Cloud key is the other option. - Blocked pages stay yours. Firecrawl Local hands pages that refuse your machine to Firecrawl Cloud. fastCRW detects block pages, escalates to a JS renderer, supports opt-in stealth renderers, and rotates across your own proxy pool, all on your side.
- JSON extraction works locally. Firecrawl Local skips the
jsonformat on local scrapes. fastCRW accepts Firecrawl's v2{"type": "json", "schema": ...}format on/v2/scrapeand runs it against the LLM you set in[extraction.llm], which can be any OpenAI-compatible endpoint, including one you host. - Field names. A few metadata keys and error messages differ slightly from Firecrawl's. Run your tests once after the swap; the migration guide lists the known differences.
Want it hosted instead?
The same Firecrawl SDK code runs against fastCRW Cloud. Set the API URL to https://compat.fastcrw.com and the API key to your crw_live_ key. New accounts get 1000 free credits, and a page costs 1 credit on every renderer.
