Firecrawl Alexandria Explained: What It Is, How It Works and What It Costs

Ask an AI agent "what did the Fed's balance sheet look like last quarter?" and it will usually run a web search, open a few news articles and paraphrase them. The real answer sits in an official data source, FRED, which web search rarely surfaces as structured data.

Alexandria is Firecrawl's attempt to fix that. Launched on September 22, 2026, it lets an agent search across official data providers, Firecrawl's own indexes and the live web in one call, then run the provider's tool directly to get structured data back.

Try Alexandria on Firecrawl ↗

What Alexandria is

Firecrawl calls it "the knowledge library for superintelligence." In practice it has three layers:

• Provider tools. Official APIs wrapped as callable tools with defined inputs, outputs and prices. Firecrawl names sources such as SEC EDGAR, FRED and the World Bank, plus licensed publishers. Its product page lists 418 providers and 2,409 capabilities (as of October 2026).
• Firecrawl indexes. Research, Developer and Government indexes covering papers, READMEs, issues and docs. Firecrawl's product page puts this at 113M+ indexed sources.
• The live web, through Firecrawl's normal search and scrape.

The core idea: web search returns pages to read, while Alexandria also returns tools to call. A tool gives you the underlying data, not someone's summary of it.

Firecrawl reports that agents using Alexandria scored 21% higher on answer quality than agents using built-in web tools, across 845 tasks with the same model and prompts and blind AI judging. That's Firecrawl's own evaluation, not an independent benchmark, but the setup is described clearly enough to be a useful signal.

How it works

There are three steps: discover → inspect → run.

1. Discover tools with search

Add alexandria to the sources of a normal Firecrawl search. You get web results and ranked tools side by side:

const result = await firecrawl.search(
  "podcast conversations about AI agents",
  {
    sources: ["web", "alexandria"],
    limit: 2,
  },
);

// Pages to read. Tools to go further.
console.log(result.web);
console.log(result.tools);

Use sources: ["alexandria"] if you only want tools. Each tool comes back with its provider, capability and description.

Two more ways to discover:

• findTools (JavaScript) or find_tools (Python) browses the catalogue with filters such as providers.
• Domain matching: when you scrape a URL with domainTools: true, Firecrawl suggests tools for that website. It's on by default in search and off by default in scrape.

2. Inspect a tool

Tool details describe the inputs (options), the response contract (response) and the execution price, so an agent can check a call is worth making before it spends credits.

3. Run it

A tool runs through the regular scrape endpoint, with an alexandria object instead of a page URL:

curl https://api.firecrawl.dev/v2/scrape \
  -H "Authorization: Bearer $FIRECRAWL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "alexandria": {
      "provider": "particle",
      "capability": "podcasts/episodes/search",
      "options": { "semantic_search": "AI agents", "limit": 2 }
    }
  }'

The result is in alexandria[0].data. Check each result for errors before using it.

From the CLI or an AI agent

The same flow is available in the Firecrawl CLI:

npm install -g firecrawl-cli
firecrawl login
firecrawl search "podcast conversations about AI agents" --sources web,alexandria --domain-tools --limit 2
firecrawl scrape particle/podcasts/episodes/search --options '{"keyword_search":"AI agents","limit":2}'

For agents, Firecrawl's MCP server exposes it through firecrawl_search (discover), firecrawl_find_tools (inspect) and firecrawl_scrape (run). So Claude, Cursor or any MCP client can use Alexandria without custom code.

What it costs

Credits come from your normal Firecrawl plan. At the time of writing that means a free tier with 1,000 credits a month, and paid plans from $19 a month (full Firecrawl review). Some providers also require an organisation admin to accept their data terms before their tools can run. The API returns a THIRD_PARTY_DATA_TERMS_REQUIRED error with a link when that's needed.

Things to know before you build on it

• Agent runs need explicit opt-in. In Firecrawl's agent endpoint, provider tools are only used when you send exchange: { "enabled": true }. Otherwise runs stick to the web. You can pin up to five providers per run with toolkits.
• Provider terms are per organisation. Providers whose terms you haven't accepted are skipped, and listed in exchange.skippedProviders.
• Prices vary by tool. Inspect the execution price before letting an agent loop over a paid tool.
• The feedback window is short. Feedback must be sent within 20 minutes of your last Alexandria call.

When to use Alexandria vs. scraping

For the third case, field-level extraction from known pages, you can also try our own extraction API (our product): describe the fields and get typed JSON with a confidence score for each one.

Verdict

Alexandria is one of the more useful ideas in agent web access this year. Instead of making scraping smarter, it skips scraping wherever an official source exists. Discovery is free, so there's little risk in adding alexandria to your search sources and seeing which tools come back for your domain. Watch per-tool pricing and provider terms before letting an agent call tools unattended.

Get started with Firecrawl Alexandria ↗

Sources: Alexandria docs, launch announcement, product page. Checked October 29, 2026. This page contains affiliate links (disclosure).