webmri

An MRI for websites and web apps. One command shows what a site publicly serves over HTTP — the stack its pages advertise, the design tokens in its CSS, its routes, forms and programmatic surface — with the evidence behind every claim. Seconds, no setup, no API keys.

webmri reads what servers send: HTML, headers and stylesheets. It does not run JavaScript, log in, or touch private APIs, so it reports the HTTP-visible surface of a site — which for most sites is the honest picture.

./webmri scan https://tailwindcss.com

webmri: 12 pages in 1235 ms, 5 stack findings

What you get

A Markdown (and JSON) anatomy of the target:

• Stack — frameworks, CMSs, hosting, CDNs, analytics, payment and monitoring services, detected by heuristic fingerprints. Each finding is labelled observed (the pattern was directly seen in a response) or inferred (a weaker pattern suggested it), with the URL and the exact signal as evidence. A fingerprint is a strong hint, never a guarantee — check the evidence.
• Design tokens — CSS custom properties from :root, most-used colors, font families, from inline styles and real stylesheets.
• Routes — the same-host pages found via the sitemap and link crawling, capped and polite. Not an exhaustive site map.
• Forms — method, target and fields of the HTML forms found in served pages, with the page each lives on.
• Programmatic surface — links and assets that look like APIs, GraphQL endpoints or API docs (inferred candidates to investigate, not confirmed endpoints).
• Performance snapshot — per-page fetch times: fastest, slowest, median, total bytes.

Install

Requires Go 1.23+ to build. The result is a single static binary with one module dependency (golang.org/x/net).

Build from a clone:

git clone https://github.com/DevelopersGlobal/webmri.git
cd webmri
go build -o webmri .

Quick start

# Scan a site (default: up to 25 pages, 8 parallel fetches)
./webmri scan https://example.com

# Save the report, and get machine-readable JSON too
./webmri scan https://example.com --out report.md --json report.json

# Just the landing page
./webmri scan https://example.com --no-crawl

# Tune the crawl
./webmri scan https://example.com --pages 100 --concurrency 16 --timeout 8s

Flags after the URL work too. ./webmri -h lists everything.

Sample output

Real excerpt from webmri scan https://tailwindcss.com --pages 12:

| name | category | confidence | evidence |
| --- | --- | --- | --- |
| Next.js | framework | observed | header x-powered-by: Next.js — https://tailwindcss.com |
| Tailwind CSS | styling | observed | asset URL .../css/... — https://tailwindcss.com |
| Vercel | hosting | observed | header x-vercel-id: sfo1::... — https://tailwindcss.com |
| Cloudflare | cdn | observed | header cf-ray: a537...-PDX — https://tailwindcss.com |

CSS custom properties (`:root`): 80 found, from --color-brand-primary to --font-plex-mono
Forms: POST https://app.convertkit.com/... on /blog — fields: email_address (email)

Measured speed

Wall-clock times from real runs on 2026-11-01 (default flags unless noted). These are this tool's own measured numbers — not a comparison with anything else. They depend on hardware, network and target; your runs will differ. Measure on your own machine before quoting them.

target	pages	wall time
example.com	1	73 ms
go.dev	12	777 ms (1.1 MB fetched)
tailwindcss.com	12	1.2 s
synthetic 200-page local site	200	26 ms (16 workers)

The whole anatomy costs one pass: every fetched page feeds stack detection, design tokens, routes, forms and API-surface extraction at once, and fetches run in parallel.

Use it with an AI agent

skills/webmri/SKILL.md is a ready-made agent skill: drop it into your agent's skills folder and the agent can run webmri, read the JSON and explain or rebuild what it finds. The JSON output is stable and self-describing, so any tool can consume it without scraping the Markdown.

Honest limits

• Static analysis of public pages only. No JavaScript execution, no logins, no private APIs, no binary analysis.
• Fingerprints are heuristic. observed means the pattern was directly seen; it still is not a guarantee. Evidence is always attached so you can check.
• Crawling is deliberately polite and bounded: same host only, capped page count, capped stylesheet and body sizes, your user-agent is identifiable by default.
• Sites rendered entirely client-side will show thin routes; their JS bundle contents are out of scope.

Contributing

Contributions are welcome — see CONTRIBUTING.md. Good first issues are labelled good first issue.

License

MIT — free for anyone, for anything.