Firecrawl CLI: Add scraping and search to agents in the terminal

Who it is for

This guide is for people who need web search, scraping, page interaction, site crawling, sitemaps, or running agent tasks from a local terminal. Officially it is written both as a command-line entry point and as an agent skill that can be installed into detected coding editors.

If your work is to read public pages first and then decide whether to change code—rather than building a scraping service yourself—this path fits better. Official notes also mention searching papers and developer materials. It is not for people who only want a few clicks in a graphical UI and cannot run terminal commands: the recommended path installs a CLI tool, completes sign-in, and writes skills into detected editors.

The official repository is

https://github.com/firecrawl/cli

Follow only the current wording on that page. Do not switch to an installer or skill directory that does not appear there.

First try in 3 minutes

Take only the single official recommended path.

Prerequisites: the terminal can run npx, and you already have a Firecrawl access method. Official notes also allow setting the FIRECRAWL_API_KEY environment variable; init with --browser uses browser sign-in. Before you run it, confirm you will sign in to your own account, and notice that the default writes to every detected coding editor, not a temporary directory.

Install location: skills are installed globally into detected editors by default. To target only one, the official form adds --agent <agent>.

Copy and run:

npx -y firecrawl-cli@latest init -y --browser

Success signal: the command finishes normally, this machine can invoke firecrawl, and the detected editors show a new skill. Do not crawl a whole site yet. Next, do a reversible check: run firecrawl setup core. Official notes say this group provides scrape, search, crawl, and interact primitives plus index-style skills; skills is an alias. After the command is accepted, ask the installed agent a read-only request, for example “Use Firecrawl to search a public document title, show the results only, and do not write files.” Seeing the results means the skill was found.

When you need application-integration skills that talk to the Firecrawl API, then run firecrawl setup build. When you need end-to-end recipes such as lead generation or deep research, then run firecrawl setup workflows. The first time, install only core so you do not write every recipe at once.

Usage examples

Example 1: Add search and scrape primitives to the current agent. The official action is firecrawl setup core, which installs the scrape, search, crawl, and interact primitives plus index skills. This fits if you already work in a detected editor and want the agent to search or scrape public pages before deciding whether to change the repo. Finish init above first, then install only core. On success, a conversation should trigger one search or scrape and show a result summary. Official materials do not give a skill-folder path, so do not invent a directory that never appears.

Example 2: You need ready-made recipes rather than single primitives. The official action is firecrawl setup workflows; the materials mention lead generation and deep research. This fits after core already works and you want recipe-style research. Confirm init succeeded, then run that command; to install into only one editor, add --agent <agent>. Success is that this skill group appears in the target editor. Recipe internals are not spelled out; existing materials only confirm that workflows is installed, and you cannot fill in unwritten sub-steps.

Limits

Existing materials stop at terminal commands, browser sign-in or FIRECRAWL_API_KEY, and the three install groups core, build, and workflows. They do not require a specific language, and they do not write “detected editors” as an exclusive list; do not infer from the default install scope that only a few clients work.

Do not apply this npx init to an environment that only allows hand-copying skill files and has no general command line. The materials also give no secondary marketplace URL or sample secret.

init and setup write skills into detected editors and have side effects. Confirm the target editor first; after running, only check whether the skills appeared—do not immediately crawl an unauthorized site. Scraping, interaction, and crawling all reach external sites; verify first with a public page you can open again.

FIRECRAWL_API_KEY is an environment-variable name, not a sample secret to paste into the body. Do not invent subcommands, skill names, or install paths that never appear. For more detail, follow the official repository:

https://github.com/firecrawl/cli