I built an email-to-URL service on Cloudflare Workers — here's what I learned about parsing email threads

Gareth Evans — Thu, 05 Mar 2026 08:57:18 +0000

The problem

Email threads are surprisingly hard to share outside email. Screenshots lose context, forwarding creates quote nesting hell, and copy-paste strips formatting.

I built ThreadCrunch — forward any email thread to share@threadcrunch.io and get a clean, readable URL back in seconds.

The stack

Everything runs on Cloudflare:

Email Workers — receive inbound email at custom addresses
Workers — API + business logic (TypeScript)
D1 — SQLite database
Pages — Vue 3 + Tailwind frontend

Total monthly infrastructure cost for a solo SaaS: basically zero until you hit scale.

The hard part: parsing email threads

Every email client formats forwarded threads differently:

Outlook uses separator lines and inline header blocks:

From: Alice
Sent: Monday, March 3, 2026
To: Bob
Subject: Re: Project update

Gmail uses "On {date}, {person} wrote:" followed by > quoted lines.

Webmail (Roundcube, etc.) uses RFC 3676 format=flowed, which joins soft-wrapped lines with trailing spaces. This completely destroys thread structure if you let your MIME parser decode it — I had to strip format=flowed from the Content-Type header before parsing and handle space-stuffing manually.

The fun edge cases:

Header fields that wrap across lines (long From/To addresses)
Outlook messages containing Gmail-style quoted replies inside them
Mixed clients in the same thread (someone replies from Outlook to a Gmail chain)

No signup wall

You don't have to sign up before forwarding. If you're not subscribed, the reply email includes a signup link and your threads get processed automatically once payment goes through.

AI analysis

There's also process@threadcrunch.io which runs the thread through Claude Sonnet for analysis — tone, intent, power dynamics, action items, and a draft reply oriented to the person who forwarded.

The key prompt design insight: doing per-message sequential analysis (tone, emotion, response timing, subtext) before synthesis produces significantly better output than asking for a flat "analyse this thread."

Developer-friendly

Every thread URL has .md and .json endpoints:

threadcrunch.io/t/{id}.json — structured JSON
threadcrunch.io/t/{id}.md — clean Markdown

Useful if you want to pipe threads into your own LLM tooling or documentation.

Try it

Demo thread: threadcrunch.io/t/5qyi25qv

$5/month, threads auto-delete after 7 days. Would love feedback from the dev community.

Forem: Gareth Evans