← Help

RSS feeds in inboxy

inboxy can poll RSS / Atom feeds for you and expose the articles through the same MCP server and REST API as your mail. Point Claude (or any MCP client) at your inboxy account and it can read your TechCrunch / AWS ML blog / Substack feeds alongside your @inboxy.net mail with one OAuth login.

Adding a feed

Open Account → Feeds, paste the feed URL, optionally label it, and click Add feed. You can also add a feed from an MCP client:

add_feed({ url: "https://aws.amazon.com/blogs/machine-learning/feed/" })

inboxy normalises the URL (lowercased host, default-port stripped, common tracking params dropped) so https://Foo.com/feed/ and https://foo.com/feed/?utm_source=x are treated as the same source.

How polling works

One poll per canonical URL — not per subscriber. If 50 inboxy users follow the same feed, the publisher sees one polite GET (with conditional If-None-Match / If-Modified-Since headers) every 60 minutes by default. Adaptive backoff doubles the interval up to 24 hours after consecutive errors; after 10 failures the feed is marked error and surfaced in the UI.

Manual Refresh is rate-limited per user (10/hour) and per source (1 / 5 min). A successful refresh benefits every subscriber.

Reading articles

Article bodies aren’t rendered in the web UI yet — for now use:

  • the MCP list_articles / get_article / summarize_article / summarize_feed tools, or
  • the REST endpoints under /v1/articles and /v1/feeds/:id/articles on api.inboxy.net.

Every <a href> in an article body is rewritten at serve time through l.inboxy.net/r/{token} — same redirector that strips trackers from forwarded mail. The publisher sees a request from inboxy’s edge, not from your IP / browser, and inboxy gets one place to defang malicious links if needed.

summarize_article and summarize_feed are client-side sampling — the LLM running in your MCP client does the work, not a server-side model. inboxy never persists generated summaries.

Privacy

Subscriptions and read state are per-user. The shared feed cache means storage stays flat as more users subscribe to the same feed, but inboxy never reveals to one user that any other user follows the same source — the subscriber count is an internal counter and is not shown in any tool, API response, or UI.

If you’re the last subscriber and you remove a feed, the shared source is reaped after a 7-day grace period along with its cached articles.

Caps

Free plans: 10 feeds. Paid plans: 50 feeds.

You can’t adjust a feed’s individual poll cadence — it’s a property of the source, shared with every other subscriber, and the adaptive backoff already keeps cadence sane.

What isn’t supported (yet)

  • Push-to-inbox: getting new articles as messages rows. Ask if you want this — it’s a one-flag-per-feed change but we’d rather wait for a real use case.
  • OPML import / export.
  • WebSub (PubSubHubbub) — every feed is polled today.
  • An in-app article reader. MCP and the REST API are the read surface.

Still need help? support@inboxy.net