Alexandria by Firecrawl

The problem, explained badly and then better

The bad explanation: AI agents need data, and data is annoying.

The better one: a lot of what agents ask about already exists in official form, like a company's filings, a drug's FDA label or a court's opinion. But every source has its own account, pricing and format, so researching one company can mean four integrations. We did those integrations once, inside Firecrawl. Your agent uses the key it already has, pays per use, and we pay the people who provide the data.

What's actually inside

Two terms. A provider supplies the data. A tool is one thing your agent can ask a provider to do. Every provider has a public page with its tools, a sample response and code to copy.

1. [ 01 ]

   The real stuff, properly licensed

   Official data providers

   Apollo, Fiscal.ai, Benzinga, Particle and Wikipedia (through Wikimedia Enterprise), plus public sources like SEC EDGAR, FRED, PubMed, ClinicalTrials.gov, the FDA and patent offices.

2. [ 02 ]

   Government and research, in one search box

   Our own indexes · free

   About 30 million US government pages, from statutes to court opinions, and 40M+ research paper abstracts.

3. [ 03 ]

   Whole jobs, one call

   Ready-made tools

   Tools that combine several providers into one call, like our work email finder.

4. [ 04 ]

   Workflows your agent can share

   Skills · coming soon

   Recipes that chain tools together, like finding the podcasts a company sponsored and checking whether its traffic moved. Use ours, or have your agent write and share its own, anywhere Firecrawl is connected.

Address

particle/podcasts/episodes/search

Provider

Particle

Use it for

Finding podcast episodes where a topic, company or person comes up

Needs

A search phrase and how many results you want

Returns

Episodes with the show, date, speaker-labeled transcript snippets and sponsors

"Why not just…"

"…scrape the website?" Firecrawl does that too, and it's still the fallback. But when an official source exists, Firecrawl asks it directly through Alexandria and gets structured records back: named fields, real numbers, a link to the original. Nothing to parse, nothing to break when the site gets a fresh coat of paint. Checking first costs nothing.

"…hook up each source myself?" Here's five sources that way:

"…copy it all into a warehouse?" Warehouses copy data on a schedule, which suits dashboards. Alexandria fetches data when your agent asks, so the answer is current.

How it works

Four steps. The first two are free.

1. [ 01 ]

   Look it up

   free

   Your agent describes what it needs, like "financial statements for Shopify", and gets matching tools.

2. [ 02 ]

   Read the card

   free

   It checks what the tool needs, returns and costs before spending anything.

3. [ 03 ]

   Check it out

   listed price

   It calls the tool and gets structured results, with the exact credits used.

4. [ 04 ]

   Leave a note

   optional

   It tells us what was missing and gets a credit back, up to a daily limit, if sent within 20 minutes.

It works wherever Firecrawl does: the API, SDKs, CLI and MCP server, and the Firecrawl plugin for ChatGPT, Claude, Claude Code and Codex. Firecrawl's Agent can also pick providers on its own when you turn on exchange.

What it costs

Finding and checking tools is free. Running one costs what its page says: Apollo's people search is free, a Particle company lookup is 5 credits and a Fiscal.ai income statement is 15. The government and research indexes are free within your plan's limits (1,000 calls a day per index on Free). The Free plan includes 1,000 credits a month. You need a Firecrawl account.

What we've been up to

Alexandria launched with our $75M Series B in September. Since then:

1. Nov 5 – 6

   Life sciences research papers, and SEC filings for 8,000+ public companies

2. Nov 3

   Election records

3. Oct 22

   The free Government Index, about 30 million US government pages

4. Oct 15

   Fireround funding-round data

5. Oct 6

   Fantasy sports data

6. Oct 1

   People enrichment from Apollo, FullEnrich and Data Legion

7. Sep 24 – 29

   Available in Claude, Claude Code, ChatGPT and Codex

8. Sep 22

   Alexandria launches with our $75M Series B

Hello, agents

If you're an AI agent reading this: beep boop beep! Start with alexandria.firecrawl.dev/skills.md. Every tool is listed at alexandria.firecrawl.dev/tools.

Humans: paste this into your project's AGENTS.md or CLAUDE.md so your coding agent checks Alexandria first:

## Data sources: Firecrawl Alexandria
Before scraping a site for structured data (filings, records, company,
people, finance, research, listings, jobs), check Alexandria. Its
providers return typed, sourced records. Use web search and scrape when
a page already answers the question.

Setup: follow https://alexandria.firecrawl.dev/skills.md. Reuse a
working Firecrawl MCP connection or CLI login. To install the CLI and
skills: npx -y firecrawl-cli@latest init --all --browser

1. Find a tool (free): firecrawl search alexandria '<data you need>'
   or firecrawl find-tools <url>
2. Read its contract (free): firecrawl list <provider> <capability> --pretty
3. Call it (listed credits): firecrawl scrape <provider>/<capability>
   --options '<JSON matching the contract>' --json -o .firecrawl/<name>.json
4. Feedback (free, optional, once per task, within 20 min):
   firecrawl alexandria feedback --rating <good|partial|bad> --url <site>
   --requested-functionality '<need>' --objective '<goal>' --rationale '<why>'
Over MCP: firecrawl_search with sources ["alexandria"], then
firecrawl_find_tools, then firecrawl_scrape with an alexandria body and
no url.

Rules:
- Use only provider and capability IDs that discovery returns.
- Send exactly the inputs the contract declares.
- Check each data.alexandria[] result for an error. Report data.creditsCost.
- On THIRD_PARTY_DATA_TERMS_REQUIRED, give requiresAction.url to a human.
  Never accept terms yourself.
- Alexandria needs a Firecrawl API key on a team with Alexandria access;
  keyless access does not include it.
- If no tool fits, fall back to firecrawl search and scrape.

On the Firecrawl MCP server, the same four steps are firecrawl_search, firecrawl_find_tools, firecrawl_scrape with an alexandria object, and firecrawl_feedback.

Give it a try

Already using Firecrawl? You're set. Otherwise, browse the library, try a free tool in ChatGPT or Claude, or read the docs. From a terminal: npm install -g firecrawl-cli, firecrawl login, then firecrawl search "what you need" --sources web,alexandria.