Workflows
Firecrawl
A web-crawling library for AI agents, licensed AGPL-3.0. The official MCP server is cataloged separately.
firecrawl/firecrawl is the main Firecrawl repository, licensed AGPL-3.0. It is the library behind search, crawl, and cleaned page content for agents. The product site is firecrawl.dev. This catalog already has the official MCP wrapper firecrawl-mcp-server; this entry is the product library and its license.
Check that you may fetch the target site, and read what AGPL-3.0 requires if you modify or distribute the software. API keys have quotas. Treat fetched text as source material. Do not republish it as your own article.
Mentioned by:
- @0x_rody mentioned this project on 14 November 2026 (26 likes and 0 reposts, captured 15 November 2026).
This page is a guide, not the original text.
What problem it solves
firecrawl/firecrawl is the main Firecrawl repository, and the product behind web search, scraping, cleanup, and interaction for AI agents. The product site is firecrawl.dev. You can call the hosted API and the official SDKs, or follow the docs and self-host the open-source server. This page is about the product itself: the REST API, the Python and Node SDKs, the CLI, and the license and capability limits that matter when you self-host.
Do not confuse it with the official MCP server, which already has its own catalog entry. MCP install steps, the keyless remote endpoint, and client config belong on Firecrawl MCP. That page covers MCP. This one covers the library, the API, and self-hosting. The two can point at each other. Do not copy the MCP tool list onto the SDK.
Who it is for
It is for people who want to call Firecrawl from their own scripts, services, or pipelines, and for teams that must run the crawling stack on their own machines. If you only want to give an assistant web tools inside Claude, Cursor, or Codex, start with Firecrawl MCP. You do not have to self-host the full server first.
Check that you may fetch the target site. The official README says Firecrawl respects robots.txt by default, and that end users are responsible for following site terms. Treat fetched text as source material. Do not republish it as your own article. API keys start with fc- and belong in an environment variable or in the header official docs require.
How to install it
Verified on 2026-11-16 (Asia/Shanghai). SDK commands come from the official Python SDK and Node SDK docs. The main repository is licensed AGPL-3.0. Official docs also say the SDKs and some UI components are MIT; follow the LICENSE file in each directory. On that day, PyPI listed both firecrawl and firecrawl-py at 4.57.0 (MIT, Python 3.8+), and npm listed firecrawl at 4.54.0 (MIT, Node.js 18+). Those version numbers do not match. This page does not force them into one number.
Official install commands are pip install firecrawl and npm install firecrawl. You can construct a client without a key, but only the methods the docs list are available, and they are rate-limited by IP. A higher quota and the remaining methods need an fc- key.
pip install firecrawl
# or: npm install firecrawl
export FIRECRAWL_API_KEY=fc-YOUR-API-KEYA minimal Python example, pointed at a page you are allowed to fetch:
from firecrawl import Firecrawl
app = Firecrawl() # keyless mode is limited to the documented methods; pass api_key for a higher quota
result = app.scrape("https://docs.firecrawl.dev", formats=["markdown"])
print("ok" if getattr(result, "markdown", None) else "empty-or-failed")Hosted plans and pricing
Plans on Pricing this review date were: Free, 1,000 credits a month, no card; Hobby, $19 a month or $16 annually; Standard, $99 / $83; Growth, $399 / $333; Scale, $749 / $599; Enterprise, custom pricing. Credits are shared. Official metering treats a basic scrape or crawl as about 1 credit per page, and Search as 2 credits per 10 results, with add-ons such as JSON costing extra. Prices can change; use it.
Self-hosting
The official self-hosting guide was pinned to release v2.11.162 on this review date. It starts the API with Docker Compose at http://localhost:3002, with authentication off. Official docs say this is not a production architecture and omits some Cloud capabilities, including screenshots and page actions in the default stack, plus Agent, Browser, and interact. Use the pinned commands, and reread the matching docker-compose.yaml before you switch releases.
A worked example
This example uses the official Python SDK to fetch Markdown from the documentation home page, as a check that install and the key (or the keyless quota) work. Use only a public page you are allowed to fetch. The call was not run here, so this page does not invent page text.
- Run
pip install firecrawl. If you already have anfc-key, put it inFIRECRAWL_API_KEY. Do not put it in the script. - Run the minimal example in the previous section, with
https://docs.firecrawl.devas the URL. - The expected effect is a response that includes a Markdown field. Open the original page and check the introduction yourself. Do not republish the body. If the call errors or asks for a key, check that you did not call a method the docs still require a key for (for example crawl, map, or agent). On this review date, the SDK docs described the keyless set as scrape, search, and interact.
- If you want MCP rather than the SDK, go to Firecrawl MCP. Note that the existing MCP guide describes the keyless path as Search, Scrape, and Parse. That is not identical to the SDK wording. Do not mix the two lists.
Common pitfalls
- Writing this entry as an MCP tutorial, or applying the MCP keyless tool list to the SDK. The two entries are split on purpose, and official wording for the keyless surface is not identical.
- Modifying or offering a self-hosted service without reading AGPL-3.0. The main repository is AGPL-3.0. The SDKs are MIT. Read the LICENSE file in each directory before you distribute.
- Exposing the Compose quickstart to the public internet. That guide disables authentication and does not add durable storage or TLS.
- Assuming self-hosting equals the full Cloud product. Official comparison notes call out screenshots, actions, Agent, and Alexandria, among other gaps. Alexandria is not included in the self-hosted server.
- Putting the API key in a repository or in chat. Keys start with
fc-and belong inFIRECRAWL_API_KEY. - Ignoring credit and failure-billing rules. The pricing page says a scrape that returns no result is not charged, but a page that comes back as 403 or 404 is still returned and costs 1 credit. Follow that page.
FAQ
How is this different from Firecrawl MCP?
This entry is the product API, SDKs, CLI, and self-hosted server in firecrawl/firecrawl. The MCP server is firecrawl/firecrawl-mcp-server, cataloged as firecrawl-mcp. If you only want to call Firecrawl from an editor, start with the MCP page.
Can I use the SDK without an API key?
Official Python and Node docs say you can construct a client without a key and use scrape, search, and interact on a keyless free tier that is rate-limited per IP. Other methods need a key. This page does not invent a fixed daily cap. Use the official rate-limit docs.
Is the SDK also AGPL-3.0?
Not necessarily. The README says the project is primarily AGPL-3.0, and that the SDKs and some UI components are MIT. Follow the LICENSE file in each directory. The official SDK packages on PyPI and npm were marked MIT on this review date.
Does self-hosting include every Cloud feature?
Do not assume that. The official self-hosting guide, pinned to v2.11.162, describes a trusted-network evaluation deployment, and the default stack omits some Cloud capabilities. If you need those capabilities, use Cloud, or verify the extra services the docs name.
How many credits does the free plan include?
The pricing page listed 1,000 credits a month and no credit card on 16 November 2026. Metering details, such as Search at 2 credits per 10 results and extra charges for JSON, belong on that page and can change.
Sources and license
The main firecrawl/firecrawl repository is primarily licensed under AGPL-3.0. SDK and related components follow the LICENSE file in each directory. This page is a paraphrase, for CQBase — Context & Query Base, of the README, docs.firecrawl.dev, the pricing page, and the self-hosting guide, checked on 16 November 2026. The most recent repository push on that day was at 08:15 Asia/Shanghai on 16 November 2026. If endpoints, the keyless surface, or prices change, the official site wins.
In this category
- Workflows
MarkItDown
Microsoft’s Python tool for turning office documents into Markdown that a model or agent can read.
microsoft - Workflows
Claude Code
Anthropic’s terminal coding agent: it reads the repo, edits files, and runs commands.
anthropics - Workflows
OpenAI Codex CLI
OpenAI’s lightweight terminal coding agent, Apache-2.0.
openai