Skip to content

docs: add the Alexandria guide and agent workflow - #1404

Draft
developersdigest wants to merge 30 commits into
mainfrom
feat/exchange
Draft

docs: add the Alexandria guide and agent workflow#1404
developersdigest wants to merge 30 commits into
mainfrom
feat/exchange

Conversation

@developersdigest

@developersdigest developersdigest commented Sep 11, 2026

Copy link
Copy Markdown
Member

Adds the Alexandria feature page with an animated agent journey and JavaScript-first request/response examples. Covers semantic discovery beside web results, reading tool contracts, provider execution through Scrape, and Find Tools progressive disclosure, with CLI and MCP examples. Places Alexandria under More and redirects /features/exchange to /features/alexandria.

The examples follow the upcoming unified tools contract and depend on the billing API update and matching client releases.

Validation: after merging current main, both feature pages and the animation compile as MDX; JSON examples, the Alexandria navigation entry, and the redirect validate. The earlier local walkthrough rendered in Mintlify and checked Search and free Find Tools through firecrawl-local. The earlier full-site Mintlify validation reported 22 existing missing-snippet references outside the Alexandria files.

Related client previews: SDKs, CLI, and MCP.

Adds an Exchange page under Core Endpoints covering discovery, invocation,
composing provider results into a web search, bring-your-own-key, pricing
and per-call errors. English navigation only; locales are managed
separately.
Exchange has no dedicated endpoint: sources: ["exchange"] finds providers
and an exchange array runs them, both on /v2/search. Adds cohort, provider
and capability filters, and drops the idempotency key from the basic
examples now that it is optional.
The example is what the request looks like; the response block does the
explaining, including that matches come back best first.
The prose already names categories, providers and capabilities; the example
only needs to show the object form.
Adds a designed-for-agents section, nests the page under Search, and drops
the errors table now that the API reference carries it.
The description strings ran to 164 characters, forcing a horizontal
scroll to read a field the page is explaining. Also drops the note about
discovery being additive.
The call example enriched a company while the discovery step above it
returned company search, so the steps read as two unrelated stories. Uses
one market-data provider end to end instead. Removes the bring-your-own-key
section until it is ready to document.
The steps each demonstrated a different thing, so the page read as
unrelated snippets. Now discovery surfaces the company and people
providers, the single call uses one of them, and the final example
composes both with a web news search into one request.

Every provider, capability, and option name is taken from the live
catalog rather than invented: the earlier examples used a market-data
provider that is not listed and option names the input schemas reject.
The page introduced features in order and left the reader to imagine why
they mattered. It now opens with a concrete brief, names what it costs to
assemble that today, and each step moves toward it: ask who can answer,
narrow, get the answer, then compose the whole thing in one request.
Frames the page on a concrete job with stakes rather than a generic
company-and-people lookup, and each step answers one part of it.

Also stops the discovery example claiming to be live ranking output. It
is trimmed to the two capabilities the page uses, and the guidance now
says matching is lexical: naming the data ranks well, describing the
situation does not.
A self-contained iframe showing an agent answering a question with one
Exchange request: it streams in character by character, counts elapsed
time while the providers run, and tallies credits as results land.

Three switchable use cases (firmographic, finance, news), each using
capabilities and prices taken from the live catalog. Sandboxed to
allow-scripts only, with no network access and no external assets.
Adds the tool-discovery stage so the agent asks the catalog what can
answer the question before calling anything, a todo list that ticks off
as it works, elapsed time and token counts while tools run, and a prompt
box under a transcript that grows upward and fades out at the top.

The running example is Firecrawl itself, and the credit total shows its
own arithmetic rather than a bare number.
Dark panel with a banded prompt row, tool calls that render as
"Calling firecrawl_exchange…" and are rewritten in place to "Called"
when they return, and a persistent status line counting elapsed time and
tokens. Turns accumulate rather than clearing, so the panel reads as one
continuing session.

Prices come back from the Exchange MCP rather than being invented: 25 per
call for crunchbase and apollo, 50 for funding rounds, 10 for the
Atlantic.
Each scenario now opens with a request that has a deliverable attached
(brief me, prepare the investor call, give me a reading list) and closes
by writing the doc. Finance blends live quotes, SEC filings, and market
news across three providers and ends in a four-company table.

The scenario labels double as the loader, filling as each one runs and
rotating on their own. Also fixes three rendering bugs: a regex that
rewrote its own markup and printed raw spans in the table header, a
cancelled scenario that kept writing into the next one's transcript, and
a box-drawing glyph whose width pushed the first result row out of
alignment.
The firmographic scenario now begins with just a contact address and
resolves the person, the company behind it, and recent coverage, closing
with an executive brief table. The table fills in quickly and the final
answer streams out fast, so the payoff lands rather than crawls.

Adds fire-flavoured spinner verbs alongside the existing ones.
The finance scenario now reads holdings.md first and then calls Exchange
for each position, so it shows providers filling gaps in data the agent
already has rather than starting from nothing. Ends in the comparison
table and an LP deck.

Coverage comes from benzinga/news-search rather than a generic web
search, in the news and firmographic scenarios too.
The scenario prompts read as clipped notes. They are now written the way
someone would actually type them, and the LP brief is wanted in ten
minutes, which is the reason to reach for one request rather than three
integrations.
Two more scenarios built on capabilities that exist in the catalog:
Coresignal workforce signals for a board update, and Wattdata plus
FullEnrich for a launch audience.

Every capability is now 5 credits per call, so the demo and the request
examples agree. Page headings describe the mechanics rather than the
narrative, so the table of contents reads as reference.
It now searches for careers pages and scrapes one before calling
Exchange, so the panel shows scrape and search working alongside
providers rather than Exchange standing on its own.

Drops Coresignal; the workforce lookup runs on Apollo instead.
The conclusion restated that some companies were hiring. It now names
who is scaling, by how much, and flags the position whose numbers
contradict the deck.
@mintlify

mintlify Bot commented Sep 11, 2026

Copy link
Copy Markdown

Preview deployment for your docs. Learn more about Mintlify Previews.

Project Status Preview Updated
firecrawl 🟢 Ready View Preview Sep 11, 2026, 2:21 PM

💡 Tip: Enable Automations to automatically generate PRs for you.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant