01
Overview
Firecrawl Alexandria is designed to let an AI agent retrieve information from more than ordinary webpages. It combines the live web with official data providers, specialized connectors and Firecrawl indexes, while exposing those sources through the Firecrawl agent tooling.
The simplest official onboarding path is Firecrawl's CLI and agent skills. Firecrawl currently tells users to run npx -y firecrawl-cli@latest init --all --browser, then restart the AI agent so the installed skills are loaded. The initializer installs the CLI, handles browser-based authentication and installs Firecrawl skills across detected compatible agents. If you already use Firecrawl MCP, the Alexandria launch also says agents can connect through Firecrawl's MCP.
Before using Alexandria, separate three concepts. A URL points to a webpage and goes through normal web scraping. An Alexandria tool address identifies a provider/capability. A Firecrawl index is a specialized retrieval source such as the Research, Developer or Government index. Alexandria is the layer that makes those non-web sources available alongside the live web.
02
Install and connect Firecrawl
The CLI's current Alexandria shorthand is important. Firecrawl documents provider/capability addresses such as benzinga/news/search or firecrawl-research-index/read. You can pass one of these addresses to firecrawl scrape in place of a URL. The explicit --alexandria form remains supported. The CLI validates a tool address as a provider/capability and does not fall back to scraping it as though it were a website.
For example, the documentation shows a provider-style call shaped like firecrawl scrape benzinga/news/search --options '{"pageSize":10}'. It also shows the Research Index form firecrawl scrape firecrawl-research-index/read --options '{"paperId":"123","query":"methodology","k":4}'. These examples illustrate the addressing pattern; use the live catalog to confirm the exact capability and required arguments for the source you actually want.
Do not guess bare provider names. Firecrawl says a bare value such as firecrawl scrape amazon fails locally and points the user toward firecrawl list or a suggested website URL. That guardrail matters because Alexandria tool addresses and ordinary URLs represent different operations. The CLI also rejects mixing URLs and Alexandria tools in the same scrape command.
03
Use Alexandria providers and indexes
A good research workflow starts by deciding which source class best matches the question. For a current announcement, use the live web. For scientific literature, inspect the Research Index. For code and developer questions, the Developer Index can surface documentation, READMEs, issues and merged pull requests. For laws or regulations, the Government Index may be more appropriate. A robust agent can combine specialized retrieval with live-web verification instead of assuming one source is sufficient.
For coding-agent workflows, install the Firecrawl skill, restart the agent, then ask it to use Firecrawl for source discovery and retrieval. Firecrawl's own launch instructions explicitly position MCP and the API as integration routes. Keep credentials out of prompts and source code; use the authentication mechanism and environment/configuration path supported by the client you are running.
Alexandria is not a reason to replace every Firecrawl primitive. If you already know the webpage URL, Scrape is usually the direct choice. If you need to discover current pages, Search is appropriate. Crawl is useful for walking a site. Alexandria becomes useful when the agent should discover or query official providers and specialized indexes in the same broader retrieval workflow.
04
Choose sources and control cost
Cost handling requires caution. Firecrawl publishes a general credit system, and its homepage says Research Index paper endpoints are free across categories, but the Alexandria launch does not define one universal standalone price for every provider and capability. Provider-specific costs can differ. Production code should therefore inspect current pricing/capability metadata and set spend controls rather than assuming all Alexandria calls have the same credit cost.
Troubleshooting: if a tool address is rejected, list available capabilities and verify the exact provider/capability path. If the agent behaves as though Alexandria does not exist after CLI setup, restart the agent so installed skills are loaded. If a command is being interpreted as a website scrape, check that you supplied a valid Alexandria tool address rather than an unqualified name. If results are incomplete, supplement the specialized source with live-web search and verify against primary sources.
Finally, treat Firecrawl's quality benchmark as directional evidence. Firecrawl reports a 21% answer-quality improvement across 845 internally evaluated tasks when Alexandria was compared with built-in web tools using the same model and prompts. That does not guarantee a 21% improvement in your workload. Evaluate your own source mix, retrieval accuracy, latency and cost before making Alexandria a production dependency.
05
Troubleshooting and evaluation
The practical pattern is simple: install or connect Firecrawl, identify the right Alexandria source or capability, call it through the supported agent tooling, combine it with live-web retrieval where useful, and preserve source attribution so the model's answer can be checked.
Sources
Primary and supporting sources
Use the linked documentation to check current pricing, access and product behaviour. The dates above describe this article; they do not guarantee that every vendor claim remains current.