feedforge

An open-source FeedBurner replacement that runs on Cloudflare's free tier. Point it at your RSS/Atom/RDF feed and it gives you a stable feed URL served from Cloudflare's edge — your origin sees one conditional GET per poll interval instead of one request per subscriber. It also does what FeedBurner never did: agents can publish short-lived feeds directly through a REST API, so a long-running task can notify your feed reader when it finishes.

• Feed proxying — parses RSS 2.0, Atom, and RDF; repairs malformed XML; emits valid RSS 2.0 or Atom (?format=atom)
• Podcasting 2.0 — enclosures, itunes fields, podcast namespace tags, and content:encoded survive normalization
• Analytics — subscriber estimates and daily reach in Analytics Engine, with a privacy-preserving daily-rotating reader hash
• Short-lived channels — capability-token REST API for agent/human coordination over RSS
• Browser view — every feed URL doubles as a subscribe-friendly landing page
• MyBrand — map your own domain to a feed
• Refresh webhook — hosts ping, feedforge re-polls immediately

Docs: https://dev.standardbeagle.com/feedforge/

Setup

1. npm install
2. Create the KV namespace: npx wrangler kv namespace create FEEDS
3. Put the returned id into wrangler.toml (kv_namespaces.id)
4. Set your Workers route/custom domain in wrangler.toml
5. npm run deploy

Managing feeds

npm run feeds -- add myblog https://example.com/rss.xml
npm run feeds -- list
npm run feeds -- remove myblog
npm run feeds -- map-domain feeds.example.org myblog

Using feeds

• Feed URL: https://<your-worker-domain>/myblog (Atom: append ?format=atom)
• Open the URL in a browser for a subscribe-friendly landing page.

Refresh webhook

Hosts can trigger an immediate refresh instead of waiting for the cron interval:

npx wrangler secret put REFRESH_TOKEN   # one-time setup

curl -X POST https://<your-worker-domain>/api/feeds/myblog/refresh \
  -H 'authorization: Bearer <REFRESH_TOKEN>'

# => {"id":"myblog","status":"ok"}

The refresh bypasses the poll interval but still sends conditional headers, so a no-change ping costs one 304. If REFRESH_TOKEN is unset the route is disabled.

Podcasts

Podcast feeds are fully supported: enclosures, itunes channel/item fields (author, image, categories, explicit, duration, episode/season/episodeType, owner, type), Podcasting 2.0 tags (guid, locked, medium, person, funding, location, value/valueRecipient, chapters, transcript), and content:encoded all survive normalization. Invalid enum values (explicit, episodeType, medium) are dropped on parse. Audio bytes are never proxied — enclosure URLs pass through to the listener's client.

Stats

Datapoints land in the feedforge Analytics Engine dataset: blobs: [feed_id, ua_class, daily_hash], doubles: [subscribers]. Example query (Cloudflare GraphQL API, analyticsEngine dataset):

SELECT blob1 AS feed, count() AS requests, sum(double1) AS reported_subscribers
FROM feedforge
WHERE timestamp > NOW() - INTERVAL '1' DAY
GROUP BY feed

Daily unique subscribers ≈ count of distinct blob3 per feed per day.

Short-lived channels (agent coordination)

Channels are feeds published to feedforge — for agent/human coordination. A long-running task creates a channel, posts items as it progresses, and a human's feed reader (or another agent) consumes the updates.

Create a channel:

curl -X POST https://<your-worker-domain>/api/channels \
  -H 'content-type: application/json' \
  -d '{"title": "Deploy bot", "ttl_hours": 24}'

# => {"id":"...","write_token":"...","feed_url":"https://.../<id>","expires_at":"..."}

Subscribe to feed_url in any feed reader. Publish an item:

curl -X POST https://<your-worker-domain>/api/channels/<id>/items \
  -H 'authorization: Bearer <write_token>' \
  -H 'content-type: application/json' \
  -d '{"title": "v42 deployed", "link": "https://ci.example/v42"}'

Delete early (channels also auto-expire after ttl_hours, default 7 days, max 30):

curl -X DELETE https://<your-worker-domain>/api/channels/<id> \
  -H 'authorization: Bearer <write_token>'

Limits: 100 items per channel (oldest dropped), 64KB per item. Reading a channel feed is public — the id is unguessable. Only the SHA-256 hash of the write token is stored.

Architecture

Cron (scheduled) polls origins with conditional GET, normalizes RSS2/Atom/RDF to valid RSS 2.0, stores in KV. fetch resolves feed id from hostname (MyBrand map) or path, records an Analytics Engine datapoint, serves XML or the HTML landing page. Origin failures keep serving the last good copy with an X-Feed-Stale header.

License

MIT