Business Brief Generator

Point it at a company's website → a branded, ready-to-read business brief.

Firecrawl MCP
Python
Claude

Walking into a meeting understanding a business takes real homework — what they do, how they're positioned, who they serve, and how they present. This does that homework from nothing but their website address: it reads the site, captures the brand, and hands back a complete, ready-to-read brief — rendered as a clean, branded PDF. Here's a real run (client details redacted):

A real, anonymised business brief produced by the generator — A real run — anonymised (name, contact and named accounts redacted)

The problem

Researching a business properly is slow, and most write-ups are either a shallow copy of the homepage or a wall of raw scraped text. The useful version is a read— what the business actually makes money on, how it's positioned, where the gaps are — and it needs to look like something you'd be happy to hand over.

The goal: give it one URLand get back a structured, accurate, nicely-presented brief — without inventing anything that isn't on the site.

Under the hood

How it's actually built. The plain version is above — everything from here down is the technical detail.

Firecrawl MCPPythonClaudeHeadless browserWAT framework

How it works

Six stages, each owned by whichever layer is right for the job — the read engine pulls the facts and the brand, the model does the synthesis, and code renders the document.

Stage	How	Why
Map	Firecrawl maps the site → the real page list	find where the content actually lives (rarely the homepage)
Read	scrape home, about, services and work → markdown	the words: what they do, the story, the team, the clients
Brand	Firecrawl 'branding' extraction → colours, fonts, logo	the full visual identity, captured automatically — no manual picking
Capture	full-page screenshot + download the brand assets locally	a visual reference, saved so the brief is self-contained
Synthesise	LLM → a structured brief (positioning, clients, team, opportunities)	the read, not a data dump — patterns, gaps, and where to help
Render	code → a styled PDF via a headless browser	a clean, branded, send-ready document every time

It captures the brand automatically

The part that makes the output feel finished: the visual identity is pulled straight from the live site — primary and secondary colours (as exact hex), the fonts, and the logo — with no manual picking. That palette in the preview above was extracted, not typed in. It means the brief can be styled in the right brand from the first pass.

Two outputs from one research pass

The research is captured once; the framing is a choice. The same underlying facts render as either of two documents:

Internal recon

A candid prep doc, neutral styling — for walking into a meeting or interview already knowing the business and where you'd add value.

Client-facing

A warm, first-person document you can hand to the business itself — opportunity-led and styled in your own brand, a deliverable that doubles as a demo of the work.

The tools

A read engine connected over MCP, plus small, reusable scripts — the deterministic execution layer of the WAT framework (Workflows · Agents · Tools).

Firecrawl MCPthe read engine — map / scrape / branding / screenshot
md_to_pdf.pyMarkdown → styled, brand-themed PDF (headless browser)
json_to_csv.pystructured extracts → clean CSV (the sibling scraper tool)

Honest limits

It reads public pages only, at a reasonable rate, and never invents facts — anything not on the site simply isn't in the brief. Founding dates, client lists and claims are taken as the company states them, flagged to verify in conversation. It's a fast, deep starting understanding — not due diligence.

Want one run on your business?

Send me a URL and I'll talk you through what it pulls — and how the architecture fits together.