Can ChatGPT or Claude read your RSS feeds and summarize them?

Can ChatGPT read your RSS feeds and summarize them every day? Only loosely. A scheduled task can run a daily prompt naming your feed URLs, but OpenAI documents neither feed reading nor a record of items already summarized. Dependable setups put a feed reader before the model: Zapier, Make or n8n, a short API script, an MCP server, or Avos.

This guide is for anyone who wants an AI to read their feeds every morning, and for developers connecting feeds to Claude or an agent. We build Avos, one of the options, so every fact about the other tools comes from their makers' own pages, linked. Where those pages are silent, we say so.

What can ChatGPT, Claude and Gemini do with a feed today?

ChatGPT

• Scheduled tasks. OpenAI's pricing page lists scheduled tasks on Go, Plus and Pro, not on Free. A task can use the plugins available to its chat (OpenAI's guide), and monitoring tasks can search the web (release notes).
• Feed URLs. No OpenAI page we found says a task fetches a URL from its prompt, or mentions RSS. Test one answer against the feed itself.
• Plugins. OpenAI's developer docs describe adding a custom MCP server from ChatGPT's Plugins page, at a public HTTPS address, subject to account and workspace policies. An RSS server added there becomes a plugin a task can use.

Claude

• Web search and fetch. With web search on, Claude can retrieve content from specific URLs you give it (Anthropic's help page), such as a pasted feed URL.
• Scheduled tasks run hourly, daily, weekly or on weekdays on all paid plans, with web research and your connectors (Anthropic's article). Feeds aren't mentioned.
• Connectors. Customize > Connectors takes a remote MCP server's URL on every plan, one custom connector on Free (Anthropic's guide). Anthropic warns about prompt injection: connect only servers you trust.
• Claude Code and the API. Claude Code's cloud scheduled tasks run a saved prompt with your connectors, at most hourly. The Messages API's web fetch tool reads a URL from the conversation for standard token costs.

Gemini

Google's scheduled actions allow up to 10 active schedules, delivered as an unread chat or a phone notification. The page doesn't mention feeds or URLs, and says answers are prepared in advance, so fast-changing information may not be current.

What goes wrong when an assistant reads a feed on a schedule?

It can miss items, repeat them, or read only the latest few.

Confirmed in the docs:

• Stale copies. Claude's web fetch tool caches results, which "might not always reflect the latest version available at the URL" unless you bypass the cache.
• Refused fetches. The same tool supports text, HTML and PDF, and can refuse a URL because of robots.txt. Test whether your feed's XML gets through.
• Blocked hosts. A Claude Code cloud task's default environment reaches only an allowlist of development domains; a feed on any other host fails with a 403 until you allow its domain, and the run can still be marked Succeeded.

Worth testing yourself:

• Repeats. No assistant documents a list of feed items already summarized. OpenAI says monitoring tasks can use information from previous runs; watch for the same item two mornings running.
• Only the latest few. Many feeds hold only their newest entries, so a failed run or a busy feed can lose items before anything reads them.
• Skimming. Given ten feeds, a model may summarize a handful. Count items in against items out.

How do you connect RSS feeds to Claude, ChatGPT or an agent reliably?

The fix for all three: let ordinary code read the feed and remember what it has seen, and give the model only new items.

Checked on 6 November 2026 against the help, docs and pricing pages of OpenAI, Anthropic, Google, Zapier, Make and n8n, linked in each section.

With an automation: Zapier, Make or n8n

Each starts from an RSS trigger, so the model never decides what's new.

• Zapier. Chain RSS by Zapier's "New Item in Feed" trigger, the Anthropic (Claude) "Send Message" action and an email or Slack step. Add Digest by Zapier to batch the day's items into one Claude call. Zapier's free plan allows only two-step Zaps, so this needs a paid one.
• Make. Pair the "Watch RSS feed items" trigger with the Anthropic Claude modules. Make's free plan has 1,000 credits a month.
• n8n. The RSS Feed Trigger polls on your schedule and the Anthropic node calls the model. On n8n's pricing, an execution is one run of a whole workflow; a self-hosted version is on GitHub.

With a short script and the API

This is the most predictable route: you decide what's new. A Python sketch with feedparser and Anthropic's SDK:

import json, pathlib, smtplib
from email.message import EmailMessage
import anthropic, feedparser

FEEDS = ["https://example.com/feed.xml", "https://example.org/rss"]
SEEN = pathlib.Path("seen.json")
seen = set(json.loads(SEEN.read_text())) if SEEN.exists() else set()

new_items = []
for url in FEEDS:
    for entry in feedparser.parse(url).entries:
        key = entry.get("id") or entry.get("link")
        if key and key not in seen:
            seen.add(key)
            new_items.append(f"{entry.get('title', '')}\n{entry.get('link', '')}\n{entry.get('summary', '')[:600]}")

if new_items:
    client = anthropic.Anthropic()  # reads ANTHROPIC_API_KEY
    response = client.messages.create(
        model="claude-opus-5-5",
        max_tokens=16000,
        messages=[{"role": "user", "content": "Write a short daily briefing from these new feed items. "
                   "Group related items, keep every link, skip anything about sports.\n\n"
                   + "\n\n".join(new_items)}],
    )
    briefing = "".join(block.text for block in response.content if block.type == "text")

    msg = EmailMessage()
    msg["Subject"], msg["From"], msg["To"] = "Your feed briefing", "me@example.com", "me@example.com"
    msg.set_content(briefing)
    with smtplib.SMTP_SSL("smtp.example.com") as smtp:
        smtp.login("me@example.com", "your-app-password")
        smtp.send_message(msg)

SEEN.write_text(json.dumps(sorted(seen)))

Run it daily with cron and keep seen.json between runs. For OpenAI, its quickstart calls client.responses.create(model=..., input=...) and reads response.output_text; the rest stays the same.

With an MCP server for RSS

The Model Context Protocol is an open standard for connecting AI apps to outside data and tools; a server exposes tools the model can call. An RSS server would offer something like "list new items in my feeds since yesterday", through a custom connector in Claude or a plugin in ChatGPT. The official reference servers include Fetch, for web content, but no RSS server; vet a community one from the MCP Registry, or write your own. A server adds access, not a schedule or memory: pair it with a scheduled task and keep the seen list in the server.

With Avos

Avos is the option with nothing to build. A News section reads any RSS or Atom feed URL you paste or pick from its feed directory, alongside other feed kinds such as X accounts and YouTube channels. Each section has its own filters, tone, format and language, and the newsletter is emailed at the time and timezone you set. Each section skips the stories it already sent you, so a quiet day reads short instead of repeating yesterday. Avos checks every RSS feed daily: a dead one is marked in the section's settings, you get one email about it if the newsletter is running, and the AI editor offers a replacement.

Avos isn't a rival model: on paid plans you pick the model an AI section writes with (GPT, Claude, Gemini, Grok and others), and Avos adds the feeds, the schedule, the design and the delivery.

How do you set it up in Avos?

1. Start a newsletter. Sign up free and describe it to the AI editor.
2. Add your feeds. In a News section's Feeds settings, paste each RSS URL or search the feed directory. Paid plans read more feeds per section. No URL? Find a site's RSS feed.
3. Write the filters in plain words, such as "only product launches and pricing changes".
4. Set the schedule: daily, on chosen weekdays or monthly, at your time and timezone.
5. Preview, then let it run. Generate an issue on demand; past issues stay in the app.

For news that can't wait, a news alert watches feeds for new items matching a condition you write. Our guide to turning RSS feeds into a daily email briefing shows what Avos does with your feeds each morning.

When is another option the better choice?

• You want to question your feeds, not receive a digest: an RSS MCP server.
• You want items in Slack: Zapier, Make and n8n deliver wherever they connect; Avos delivers by email and in the app.
• You want custom processing, such as tagging items into a database: the script.
• You already pay for ChatGPT or Claude and follow a few quiet feeds: try a scheduled task for a week, with our ChatGPT daily briefing guide. Avos vs ChatGPT and Avos vs Claude compare both with a dedicated briefing.

FAQ

Can ChatGPT summarize my RSS feeds every day for free?

Not on a schedule: OpenAI's pricing page lists scheduled tasks on Go, Plus and Pro, not on Free (checked on 6 November 2026). Its pages also don't document a task reading feeds, so test it on a paid plan. Avos's free plan reads RSS feeds and emails a briefing on your schedule.

Can Claude read an RSS feed?

Yes, when you give it the URL with web search on. For a daily read, use a scheduled task on a paid plan, a custom connector to an RSS MCP server, or the API's web fetch tool.

What is the best way to connect RSS feeds to an AI agent?

Let code read the feed, remember what it has seen and send only new items to the model: an RSS trigger in Zapier, Make or n8n, a short script, or an MCP server with its own seen list. Asking the model to fetch the feed itself is the least predictable route.

Does Avos use ChatGPT or Claude?

It can. On paid plans you choose the model each AI section writes with, including GPT, Claude, Gemini and Grok. Avos adds the feed reading, the schedule, the design and the email.