Replacing My Feed With Claude

i've been experimenting with using Claude to replace most of my OpenClaw usage, and it's gotten to the point where i don't really need anything else. the setup is a Claude scheduled task that runs on a cron, generating and curating content for me every hour. because Claude can run MCP servers, interact with the browser, and control my Mac directly, it effectively acts as its own OpenClaw system.

the problem with feeds

instead of brain-scrolling or doom-scrolling through information i'm never going to use, i subscribe to sources i actually care about and let Claude intelligently search through the results. i can dislike a post, save a different one, like another, and Claude queries these signals over time. because i'm leaving the decision layer to Claude inside a scheduled task, it catches stories i would've otherwise missed. it doesn't just surface new stories, it understands why they matter based on my history.

the setup

the core of this is a Claude scheduled task called "Content gen" that runs hourly. it has access to the browser, Cloudflare, and bioRxiv through MCP servers, and it generates content for what i call the "brain" feed app, a personal infinite scroll of tweet-length posts.

the Claude scheduled task config. runs hourly, has access to browser, Cloudflare, and bioRxiv.

each run, it generates candidate posts, judges them for relevance, and inserts the ones that pass into a D1 database. here's what a typical run looks like:

a run summary. 7 candidates generated, 5 survived judging, 5 inserted. each post gets a priority score.

adaptive awareness

the interesting part is giving Claude permission to modify its own system prompt. for something like security or geopolitics, you want the system to be adaptive. when the Iran conflict started escalating recently, Claude picked up on it and started prioritizing related stories without me telling it to. it gives warnings for things that actually matter and that affect me directly.

plugging in your own data

you can plug it into basically anything. i have it connected to my mail system so it can locally check for relevant information. one of the more useful integrations is my Twitter bookmarks. every few hours it checks them automatically and looks for signals. say i bookmark a post about the Iran conflict, that gives Claude a hint to start surfacing and updating me about that topic. it becomes an actual information point that most other platforms would miss.

beyond that, i subscribe to DeepMind, Google's research blog, Uber's engineering blog, Yahoo, various AI research blogs, OpenAI's blog, and you can plug in your own data sources too. papers from bioRxiv, preprints, whatever you want.

the end result. hourly summaries with priority scoring, personalized to what i actually care about.

why this is different

the result is hourly summaries that don't really have any parallel in terms of content curation. the level of personalization you get from plugging in your own signals, your bookmarks, your likes, your mail, your subscriptions, means the feed is genuinely yours. no algorithm optimizing for engagement, no ads, just information filtered through something that actually understands what you care about.