Build with AI

If you're developing with AI, Firecrawl offers several resources to improve your experience. Firecrawl ships with skills — self-contained knowledge packs that AI coding agents discover and use automatically. One install command gives agents CLI skills for live web work and workflow skills for producing repeatable deliverables. Build skills for integrating Firecrawl into application code are a separate step (firecrawl setup build). Firecrawl users can get an API key in two ways. See Get credentials below.

• Get credentials
• CLI and agent skills
• Using Firecrawl as a Tool
• Agentic Debugging
• Firecrawl MCP Server
• Firecrawl Docs for Agents
• Quick Start Guides
• Agent Harnesses
• SDKs

Get credentials#

Firecrawl users can get an API key in two ways. Most users should sign in through the dashboard or CLI. If your agent platform supports WorkOS ID-JAG, it can register directly with Firecrawl.

Which should I use? Use the dashboard or CLI unless you know your platform supports WorkOS ID-JAG.

How you might arrive:

• Docs: you landed here. Pick the option below that matches how you can sign in.
• API 401: ID-JAG capable agents can follow the discovery flow in auth.md Step 1. Everyone else should use the dashboard or CLI.
• Direct URL: follow SKILL.md for browser sign-in and setup, or auth.md for WorkOS ID-JAG registration.

Can't sign in or get a key? The CLI, SDKs, and REST API allow keyless search, scrape, interact, and parse. Hosted MCP exposes the narrower keyless Search, Scrape, and Parse surface; interact requires an account connection or API key there. Keyless access is rate-limited and meant as a fallback, so move to a free account or API key as soon as one is available. See Keyless (no API key) for details.

Once you have an API key, continue with CLI and agent skills below.

CLI and agent skills#

The Firecrawl CLI lets your agent search, scrape, interact, crawl, map, extract, and run agent jobs from the terminal. It's built for humans, AI agents, and CI/CD pipelines.

The Firecrawl skills are self-contained knowledge packs that AI coding agents like Codex, Claude Code, Cursor, and OpenCode discover and use automatically. A single install command sets up the CLI tools for live web work and the workflow skills for producing repeatable deliverables. Build skills for integrating Firecrawl into application code are installed with firecrawl setup build:

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

• --all skips agent selection and initializes every detected agent
• --browser opens the browser for Firecrawl authentication automatically
• --skip-auth, --skip-install, --skip-skills, and --agent <name> are also available. See firecrawl init --help.

After install, verify everything is working:

firecrawl --statusfirecrawl scrape "https://firecrawl.dev"

To reinstall or scope to a specific agent later:

firecrawl setup skills      # CLI / core skillsfirecrawl setup build       # build skillsfirecrawl setup workflows   # workflow skills

What the install gives you#

Firecrawl skills cover three categories. init installs CLI and workflow skills. Run firecrawl setup build for build skills. Each segment lives in its own repo so it can evolve independently:

• firecrawl/cli — CLI skills for live web work
• firecrawl/skills — build skills for app integration
• firecrawl/firecrawl-workflows — workflow skills for repeatable deliverables

CLI skills — for live web work during an agent session:

The install above includes both index skills. To add just the paper-search one to an existing setup — worth doing when the work is biomedical or scientific literature — run:

npx skills add firecrawl/skills@firecrawl-research-index

Build skills — for integrating Firecrawl into application code:

Workflow skills — outcome-focused skills that produce a concrete deliverable from Firecrawl web data:

Choose your path#

CLI and workflow skills come from the install above. Build skills are a separate step: firecrawl setup build. The difference is what happens next:

# Search the webfirecrawl search "best open-source web crawlers"# Scrape a page into clean markdownfirecrawl scrape https://docs.firecrawl.dev# Crawl a whole sitefirecrawl crawl https://docs.firecrawl.dev

# Multi-source research brief on a topic"Use firecrawl-deep-research to write a brief on AI agent frameworks"# Pre-meeting intelligence for a sales call"Use firecrawl-lead-research to brief me on stripe.com before my 3pm call"# On-page SEO audit with prioritized fixes"Use firecrawl-seo-audit on https://example.com"# Clone a site's design system into DESIGN.md"Use firecrawl-website-design-clone on https://linear.app"

The full onboarding definitions live at:

• Browser sign-in, CLI setup, skills, MCP, and dashboard keys: firecrawl.dev/agent-onboarding/SKILL.md
• Agent registration with WorkOS ID-JAG: firecrawl.dev/auth.md

Agents can fetch either doc directly: use agent-onboarding/SKILL.md for human-in-the-loop CLI/browser onboarding, or auth.md for WorkOS ID-JAG registration.

Using Firecrawl as a Tool#

Firecrawl gives agents five core tools for working with the web. Each tool maps to an API endpoint and a CLI command. Agents pick the right tool based on what they need:

# CLIfirecrawl search "latest OpenAI API pricing"

# REST APIcurl -X POST https://api.firecrawl.dev/v2/search \  -H "Authorization: Bearer fc-YOUR_API_KEY" \  -H "Content-Type: application/json" \  -d '{"query": "latest OpenAI API pricing"}'

# CLIfirecrawl scrape https://docs.stripe.com/api/charges

# REST APIcurl -X POST https://api.firecrawl.dev/v2/scrape \  -H "Authorization: Bearer fc-YOUR_API_KEY" \  -H "Content-Type: application/json" \  -d '{"url": "https://docs.stripe.com/api/charges"}'

# CLIfirecrawl crawl https://docs.firecrawl.dev --limit 50

# REST APIcurl -X POST https://api.firecrawl.dev/v2/crawl \  -H "Authorization: Bearer fc-YOUR_API_KEY" \  -H "Content-Type: application/json" \  -d '{"url": "https://docs.firecrawl.dev", "limit": 50}'

# CLIfirecrawl map https://docs.firecrawl.dev

# REST APIcurl -X POST https://api.firecrawl.dev/v2/map \  -H "Authorization: Bearer fc-YOUR_API_KEY" \  -H "Content-Type: application/json" \  -d '{"url": "https://docs.firecrawl.dev"}'

# CLIfirecrawl scrape https://example.comfirecrawl interact "Click the pricing tab and extract the plan names"

# REST API# scrapeId comes from the scrape response (data.metadata.scrapeId)curl -X POST https://api.firecrawl.dev/v2/scrape/SCRAPE_ID/interact \  -H "Authorization: Bearer fc-YOUR_API_KEY" \  -H "Content-Type: application/json" \  -d '{"prompt": "Click the pricing tab and extract the plan names"}'

How agents chain tools together#

Most agent workflows combine multiple tools. A typical pattern:

1. Search to find relevant pages → get a list of URLs
2. Scrape the most relevant URLs → get clean content
3. Interact when the scraped page needs follow-up actions
4. Agent when the task needs autonomous discovery or structured multi-page extraction

For bulk work, agents use Map to discover URLs first, then Crawl or selectively Scrape the pages they need.

Agentic Debugging#

When a Firecrawl call fails or returns unexpected results, your agent doesn't have to escalate to a human. The /support/ask endpoint is an AI support agent built for agent-to-agent communication — it diagnoses issues with your jobs, account, and API usage, then returns a verified answer with machine-readable fix parameters your agent can apply directly.

Wire it into your agent's error-handling flow so it can self-recover from scraping failures, crawl issues, and configuration problems — typically in 15–30 seconds, no human in the loop.

How it works#

1. Your agent describes the problem — a natural-language question describing the issue.
2. The support agent investigates — it inspects job logs, account state, documentation, and source code.
3. The support agent validates — when possible, it tests a fix against the live Firecrawl API (e.g., retrying a scrape with adjusted parameters).
4. Your agent gets a verified answer — a prose answer, machine-readable fixParameters to apply directly, and validation results showing whether the fix was tested.

Example#

Send a question, plus an optional rationale to give the support agent context about what your end user is trying to accomplish:

curl -X POST https://api.firecrawl.dev/v2/support/ask \  -H "Authorization: Bearer fc-YOUR_API_KEY" \  -H "Content-Type: application/json" \  -d '{    "question": "my crawl returned 3 pages but I expected 50",    "rationale": "user is on their third failed crawl attempt today"  }'

The response includes an answer, a confidence rating, optional fixParameters (e.g., {"waitFor": 5000}) your agent can pass to the next call, and validation showing whether the fix was tested against the live API.

Firecrawl MCP Server#

MCP is an open protocol that lets compatible applications give models access to tools. The open-source Firecrawl MCP server supports search, scraping, browsing, crawling, mapping, structured extraction through Scrape JSON, and autonomous Agent jobs.

Firecrawl Docs for Agents#

You can give your agent current Firecrawl docs in a context-aware way. Agents can self-onboard by pulling these resources directly — no human wiring required.

Docs for this page: https://docs.firecrawl.dev/ai-onboarding.md

Here are the Firecrawl docs: https://docs.firecrawl.dev/llms-full.txt

Quick Start Guides#

Drop-in quickstarts for the stacks agents build on most often. Point your agent at any of these to scaffold a working Firecrawl integration end-to-end.

Prefer to let Cursor drive? One-click install the Firecrawl MCP server and start prompting in Cursor:

See the full list of quickstarts (Express, NestJS, Fastify, Hono, Bun, Remix, Nuxt, SvelteKit, Astro, Mastra, Django, Flask, Elixir, Java, Spring Boot, .NET, ASP.NET Core, and more) in the left sidebar.

Agent Harnesses#

Firecrawl works with the runtimes and frameworks agents actually live inside — coding agents, agent SDKs, and model aggregators. Most coding harnesses can auto-discover the Firecrawl skills via npx -y firecrawl-cli@latest init --all --browser; the rest call Firecrawl as a tool over MCP or the REST API.

SDKs#

Official, typed SDKs covering the full Firecrawl API surface. Point your agent at the language matching your stack.

Firecrawl also has first-class SDK bindings for the major LLM SDKs and agent frameworks — see LLM SDKs and Frameworks for OpenAI, Anthropic, Gemini, Google ADK, Vercel AI SDK, LangChain, LangGraph, LlamaIndex, Mastra, and ElevenAgents.