Ask ChatGPT about a company’s latest quarterly filing and you usually get a summary of a page that discusses the filing — not the filing itself. The primary record sits behind a provider API the model has no way to call. On September 29, 2026, Firecrawl announced that its Alexandria service is now available inside ChatGPT and Codex through the official Firecrawl plugin, giving agents a direct route to those structured sources.
From secondhand summaries to primary records
Alexandria works under agreement with more than 100 data providers, plus Firecrawl’s own Research, Developer, and Government indexes. When you ask a question, the plugin searches for a matching tool, shows you its inputs and listed price, then runs it. If no provider fits, it falls back to ordinary web search and scraping, so coverage gaps don’t dead-end the request.
The practical difference is what the agent reads. Instead of whatever a blog post happened to quote, it gets the provider’s own record — quarterly filings, government spending data, research papers, package registries, real estate listings, podcast transcripts.
Firecrawl reports an evaluation across 845 tasks with the same model, prompts, and blind AI judging: agents using Alexandria scored 21% higher on answer quality than agents using built-in web tools. Treat that as the vendor’s own eval, but the direction matches the design — primary sources beat paraphrases when the question turns on a specific number.
How it behaves in practice
The flow is three steps: find a tool, check what it needs, run it. A query like “find remote software engineering jobs posted this week, with salaries and application links” is a bad fit for plain search because job boards expose structured data through APIs. The plugin picks a provider, calls it, and shows you what was chosen and what the data cost.
You can also scout before spending anything. Ask it to list the providers covering federal contract awards along with per-call prices, and discovery is free. Billing is straightforward: inspecting tools costs nothing, running one charges its listed price.
One operational detail worth flagging for teams: some providers require accepting their terms before data can be pulled. Acceptance binds your whole organization, so by default an admin accepts terms in the Firecrawl dashboard. If your org enables agent acceptance, ChatGPT can accept after displaying the terms. Either way, you need a Firecrawl account connected — keyless sessions can’t use Alexandria at all.
The Developer Index is the quiet highlight for Codex users
Codex gets something more targeted: the Developer Index, built over public repositories, GitHub issues, merged pull requests, READMEs, and code documentation. Anyone who has asked an agent “how do I do X with library Y instead of Z” knows the real answer often lives in a merged fix or an issue thread, not the docs. The index searches that material directly and hands Codex the matching passages.
This connects to a theme from Alexandria’s earlier launch, where source selection itself became an API call rather than a manual decision. Now that same selection layer runs inside the tools developers already have open.
What it changes for your builds
If your product or workflow depends on agents answering from primary data — filings, contracts, registry records — this removes the workaround of scraping blog summaries or wiring provider APIs by hand. The tradeoff to keep in mind is cost visibility: discovery is free, but every tool run is billed, so agents that loop or retry against paid providers can add up. Start with price-listing queries before letting an agent execute freely.
The limitation is coverage. Alexandria only helps where a provider or index exists; outside that catalog, you’re back to the open web. Worth a quick test if you ship agent features: install the plugin, ask for a filing you can verify, and see which provider it reaches for.
Sources
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.