Setting a fantasy lineup is a small data-integration problem most people solve by hand: rankings on one site, projections in a different scoring format on another, injury reports on a third, and none of it knows your roster. Firecrawl’s new fantasy football pack, announced October 6, 2026, turns that chore into agent calls — and it’s a useful template for how structured data gets sold to agents generally.
What actually shipped
Per the Firecrawl announcement, fantasy football data is live in Firecrawl Alexandria, the company’s capability catalog that already works from Claude, ChatGPT, Codex, and other agent tools. Your agent can read your league (via a Sleeper username, an ESPN or Yahoo league link, or a roster screenshot), pull the week’s projections in your league’s scoring format, check injury reports, and recommend start/sit decisions with a one-line rationale plus top free-agent adds.
The pack bundles several distinct capabilities: weekly projections and rankings from six expert sources across PPR, half-PPR, and standard scoring; projections derived from sportsbook player props; season and rest-of-season boards including dynasty and ADP data; injuries and player news; league schedules, rosters, and standings; and accuracy checks comparing last week’s projected versus actual points.
The part worth stealing for your own product
Two design choices here matter more than football.
First, the lineup tool merges all six projection sources by player and shows the consensus next to the range. A narrow range means the sources agree; a wide range flags a close call, so the tool also grades each opponent by fantasy points allowed to that position. That’s a good pattern for any multi-source data layer: don’t average away the disagreement — surface it as a signal.
Second, discovery and pricing are structured. Every capability lists its inputs, response shape, and price before the call, and the RSS summary notes discovery is free. The agent can figure out how to call a source — and what it costs — before spending anything. One Firecrawl key pays for everything in credits; no per-site keys, no scraping.
This echoes a point from an earlier post on place data agents can act on: the hard part isn’t fetching data, it’s packaging it so an agent can select it, call it correctly, and reason about what it got back. Firecrawl’s fantasy pack is that packaging applied to a consumer use case.
How you’d actually use it
The announcement frames each workflow as a single natural-language request from ChatGPT, Claude, or Codex:
- Settle a start/sit call by comparing players across every source’s ranking, then checking the injury report.
- Work the waiver wire against rest-of-season projections and current injuries.
- Evaluate a trade using season-long projections and remaining schedules.
- Audit accuracy by pulling projected-versus-actual from last week.
- Prep a draft with ADP broken out by scoring format and league size.
You don’t need to know capability names. Describe the data you want and Alexandria returns matching capabilities, ranked, with inputs and prices. Discovery also works from any MCP-connected tool or the CLI.
Scope and limits
The coverage goes beyond football — the sports category spans NFL, NBA, MLB, NHL, college sports, and soccer, plus prediction markets and ticket prices, with new providers weekly across 26 categories. Two caveats from the source: not every projection source covers every position or scoring format, and betting-line projections are only as good as the consensus lines behind them. If you build on this, treat the source range as part of the answer, not noise to hide.
The practical takeaway: if you’re exposing data to agents, study the shape here — self-describing capabilities, visible prices, merged consensus with ranges, and a single billing path. That combination is what makes a data source something an agent can choose on its own.