From a search result to a verified record.
Four steps stand between an article and a row in the feed, and each one runs on a Firecrawl primitive you can call yourself.
- 01[ Search ]
Search finds the articles
Eight queries per preset across news and the web. A result that cannot be an article, like a site root, a tag page, or an investor directory, is declined before any scrape.
search · news + web
- 02[ Scrape ]
Scrape reads each into a schema
One scrape per page with a JSON format: company, stage, amount, date, and investors, plus the sentence that states them. A round whose sentence is not on the page is dropped.
scrape · json format
- 03[ Verify ]
Every record is read back
Before a record publishes, a model reads it against every source together and cites the page and sentence for each field. Without a model key, the judge reads the primary page with Firecrawl.
reconcile · every source
- 04[ Enrich ]
Enrichment adds the company
The company's homepage, checked against the record first, gives the profile and the logo through the branding format. Alexandria adds sourced facts on the verified domain.
branding · alexandria
One store, every way to read it.
The dashboard, the feeds, the API, and the MCP server all read the same verified records, so what an agent gets is what you see.
The feed. Every round with its logo, stage, amount, lead, and a Verified mark with its source count. Open a row for the evidence.
Companies and investors. A page for each: every round on record, total raised, the syndicate, co-investors, and Alexandria facts.
The API. JSON, CSV, and JSONL over /api/v1, with cursor pagination and the same filters as the feed. Described in OpenAPI.
The MCP server. Five tools at /api/mcp, so an agent can list, search, and read rounds the way the dashboard does.
Delivery. RSS, Atom, and JSON Feed, Slack in one click, Standard Webhooks, and an email digest, each from one outbox.
Share cards. A company or investor link unfurls in Slack with its logo, latest round, and total raised.
Ask it from a terminal, or hand it to an agent.
The same records over HTTP and over MCP. The response below is a row from this instance, trimmed to a few fields.
About 600 credits a week to run your own.
Coverage is one knob with four rungs. Every rung is the same product; the credits scale with how wide and how often it reads. Settings prices each rung against your presets before you switch.
- Lite[ Default ]
A weekly sweep of the news, homepage profiles and logos
2.4K to 2.6Kcredits a month
- Standard
Daily, news and web, corroborated, with Y Combinator facts
22K to 24Kcredits a month
- Deep
Twice daily, publisher sections and Form D, automatic research
168K to 172Kcredits a month
- Max
Hourly everything, a dossier on every company
1.3M pluscredits a month
A fresh store opens on its first sweep: last week of Series A and B, about ten minutes, about 560 to 620 credits. Runs shows the credits measured beside every estimate.
Get a Firecrawl API keyYour own feed, four commands away.
No database to set up: without a database URL it runs on embedded Postgres under ./data. On Vercel the button provisions Neon, and cron runs the sweeps.
- License
- MIT
- Runtime
- Node 24+ and pnpm
- Database
- Embedded Postgres, or any Postgres 15+
- Schedule
- Vercel cron, or pnpm start ticks itself
git clone https://github.com/firecrawl/fireround.git && cd fireroundpnpm installcp .env.example .env # paste your Firecrawl keypnpm dev # http://localhost:3000FIRECRAWL_API_KEY=fc-... # the one you needDASHBOARD_PASSWORD=... # optional: keeps a private instance privateCRON_SECRET=... # any random string, 16+ charactersEvery other variable is optional; the README lists them. Configuration
The feed is live.
Every round the web reports, with the pages that reported it.
