Firecrawl Research Pipeline

Crawl, search, and ground a research brief from a handful of seed URLs.

This is one of FrootAI's cross-server composition recipes — a single solution play whose agent attaches Firecrawl + Tavily + Context7 and runs them as one pipeline. Firecrawl captures the pages, Tavily broadens the search, and Context7 grounds every claim in real library docs. The agent synthesizes a cited brief instead of a hallucinated one.

What you'll build

The agent's task, in one line:

Crawl the target site with Firecrawl, enrich findings via Tavily search, ground them with Context7 docs, and assemble a research brief.

It attaches firecrawl, tavily-ai, context7 via the play's mcp_scope, runs the loop end to end, and detaches when it's done.

Why compose instead of chain

Federating MCP servers behind one agent means the model decides which tool to call when — no brittle glue code wiring outputs to inputs. The attach list is declarative; the engine wires the servers before the first turn and tears them down after. You get the composition without owning the plumbing.

Run it

The recipe ships a runnable, offline harness plus the full mcp_scope.attached snippet, a cost breakdown (~$17.80 per 100 runs), and a security note for every credential. Read the full recipe:

- 📖 Cookbook recipe - ▶️ Open in Studio

The takeaway

The interesting MCP work isn't any single server — it's the federation. This recipe is one worked example of many in the FrootAI cookbook. Compose, don't chain.