Case Study: Anthropic builds a parallel web research agent with Firecrawl

Anthropic builds AI agents in minutes with Firecrawl’s parallel subagents and 1,000-credit monthly free tier

Anthropic, the AI research and deployment company, needed to build a non-interactive, automated agent for auditing Python package dependencies. Their challenge was to move beyond their interactive Claude Code CLI tool to create a script that could run autonomously, requiring web research capabilities to gather data from PyPI and GitHub for a structured risk assessment.

Firecrawl provided the solution through its MCP server, which was integrated with the Claude Agent SDK. This allowed Anthropic to build an orchestrator agent that dispatched parallel researcher subagents; each subagent used Firecrawl to scrape clean, structured data from web pages. The result was an automated dependency auditor that produced a color-coded risk report, significantly reducing research time by running audits in parallel and costing roughly $0.50-$1.50 per audit run.

Anthropic