Firecrawl Web Agent

Firecrawl runs a research-grade autonomous agent at firecrawl.dev/app/agent, powered by Spark 1 models optimized for structured web research. This repo gives you the open-source foundation to build your own — fork it, swap models, add skills, and deploy however you want.

Get started

# 1. Install the Firecrawl CLI and authenticate
npx -y firecrawl-cli@latest init -y --browser

# 2. Scaffold an agent project
firecrawl create agent -t next

Open Source

Each layer builds on the one below it. Start at the top for a ready-to-use app, or go lower in the stack for finer control over the primitives.

Examples

How it works

The agent combines web tools with an AI model in a loop — it plans, acts, observes, and repeats until the task is done. The harness is Deep Agents (from LangChain), which gives us the plan-act loop, parallel task sub-agent spawning, and on-demand SKILL.md loading out of the box. Our agent-core wires Firecrawl's tools into that runtime and layers on structured output and streaming.

• Harness — Deep Agents. Provides the agent loop, sub-agent spawning, skills loading, and context management.
• Tools — Search, Scrape, Interact (browser automation), bash. Powered by firecrawl-aisdk.
• Skills — reusable SKILL.md playbooks. Auto-discovered from agent-core/src/skills/definitions/, loaded on demand via Deep Agents' skills middleware.
• Subagents — parallel workers for independent tasks, spawned via Deep Agents' task tool. Each has its own tool set and session state (e.g. an isolated interact browser session).
• Output — structured results via formatOutput (JSON) and data processing via bashExec, a set of bash tools powered by just-bash.

Project structure

License

MIT