Skip to main content
Back to docs
workflowcrawl4ai
Install
Source: packages/harness-kit/src/registry/bundles/workflow/crawl4ai/README.md

Crawl4AI

Open-source web scraping with JavaScript rendering, LLM extraction, deep crawl, and YouTube/PDF processing — self-hosted, no API key required.

What it installs

ArtifactPath (in your project)Purpose
MCP server.mcp.json → crawl4ai19 tools via walksoda/crawl-mcp
Skill.agents/skills/crawl4ai/Tool selection guide, LLM extraction, context overflow prevention

Setup

Requires Python 3 and uv:

pip install uv   # or: curl -LsSf https://astral.sh/uv/install.sh | sh

The MCP server installs automatically on first use via uvx.

Key tools

CategoryTools
Single pagecrawl_url, extract_structured, extract_entities, extract_schema
Multi-pagecrawl_deep, crawl_multiple
YouTubeyoutube_transcript, youtube_metadata, youtube_batch
FilesPDF, Word, Excel, PowerPoint → markdown

LLM extraction

extract_structured uses an LLM to extract semantically — useful when data is scattered or requires reasoning rather than CSS selection. Set extraction_type: schema and provide instructions.

Avoiding context overflow

Pass output_path on large crawls to write results to disk instead of returning inline:

tool: crawl_deep, seed_url: ..., output_path: /tmp/result.md

The tool returns a summary + file path. Read specific sections as needed.

vs Firecrawl

crawl4aiFirecrawl
CostFree (self-hosted)$16–$83/month SaaS
SetupMedium (needs Python/uv)Zero (just API key)
PrivacyOn-premiseCloud
YouTube/PDFYesNo
ScaleUnlimitedTier-limited

Choose crawl4ai for high-volume, privacy-sensitive, or Python-native workflows.
Choose Firecrawl for zero-setup, quick HTML→markdown scraping.

Self-hosted Docker (optional)

For production or high-volume use:

docker run -d -p 11235:11235 --shm-size=1g unclecode/crawl4ai:latest

Connect via MCP SSE at http://localhost:11235/mcp/sse.

Pairs well with

  • firecrawl — complementary: Firecrawl for quick scrapes, crawl4ai for deep/structured/video extraction
  • brave-search — find URLs to crawl, then extract with crawl4ai
  • tavily — Tavily for search + AI answers, crawl4ai for raw page content