Use cases

Give your AI agents fresh web content

Agents are only as current as what they can read. Owlie watches the sources you choose, extracts clean text from every new item, shapes it the way your agent expects, and hands it over by webhook or MCP. No scrapers or transcription pipelines to maintain.

The recipe

One Operation is the whole pipeline. These are the settings to choose when you create it; you can change any of them later.

  1. Sources

    Add the sources your agent should know about: YouTube channels, RSS feeds, podcasts and subreddits, or a single video or article. Owlie checks the feeds every hour.

  2. Filter

    Choose Everything to pass every new item on, or keep your agent’s context focused with keywords or AI criteria, so it only receives what is relevant to its job.

  3. Transform

    Choose Extract only to pass clean text, Analyze with AI for a written brief, or Structured content for JSON with your own fields that your code can rely on.

  4. Deliver

    Add a webhook to receive each Result as a signed JSON event the moment it is ready. Agents can also read your Results over MCP with a personal access token.

Example

Illustrative example with a made-up release note. The fields shape the data; the webhook event wraps it with the item, its Source and why it passed the filter.

  1. Extracted content

    Clean text Owlie extracted from a new post in a product’s changelog feed.

    Article

    Acme API 2026-10: rate limit changes

From November 1, the free tier allows 60 requests per minute instead of 100. Paid plans are unchanged. Requests over the limit now return 429 with a Retry-After header. The /v1/search endpoint is deprecated and will be removed on January 31; use /v2/search instead.

  2. Fields

    Each field says what to extract, so your agent always gets the same keys.

    Fields: what to extract

    {
  "summary": {
    "type": "text",
    "description":
      "The change, in one sentence"
  },
  "effective_date": {
    "type": "date",
    "description":
      "When the change takes effect"
  },
  "action_required": {
    "type": "boolean",
    "description":
      "Must API users change code?"
  },
  "deprecated_endpoints": {
    "type": "list of text",
    "description":
      "Endpoints being removed"
  }
}

  3. Webhook event

    Sent to your URL with a signature header. Shortened here.

    POST result.created

    {
  "event": "result.created",
  "version": 1,
  "resultType": "structured",
  "data": {
    "summary":
      "Free tier limit cut to 60/min",
    "effective_date": "2026-11-01",
    "action_required": true,
    "deprecated_endpoints": [
      "/v1/search"
    ]
  },
  "resultUrl": "https://owlie.io/…",
  "meta": {
    "source": { "type": "rss", … },
    "item": { "title": "…",
      "url": "…", "content": "…" },
    "match": null
  }
}

Variations

The same recipe with different Sources and fields.

  • Keep a knowledge base current

    Choose Extract only and send every new article, transcript and post to a webhook that stores it in your search index or vector database, with its title, Source and original link.

  • An agent that reads your Inbox

    Connect an AI assistant to Owlie over MCP. It can list your newest Results, filter them by Label or text, and open one in full to answer questions about what changed this week.

  • Trigger automations from new content

    Send structured Results to a workflow tool. A boolean field such as “action required” decides whether to open a ticket, post a message or do nothing.

  • Let an agent set up its own sources

    With a token that allows creating Operations, an agent can start following new Sources over MCP, with the same settings as the Operation form.

Good to know

What this recipe can and can’t do today.

  • Webhooks are delivered once

    Owlie sends each event once and waits up to five seconds for your reply. A failed delivery is not retried, but the Result stays in your Inbox and the Operation shows the latest delivery status.

  • Verify the signature

    Each webhook has one signing secret, shared by every Operation that sends to it, and each request is signed with it. Check the signature header before you trust the payload.

  • MCP reads Results, not the live web

    Over MCP, agents read the Results your Operations have already produced. To get something new, an agent creates an Operation and reads its Results once they are ready.

  • What it costs

    Extract only costs 1 credit per Result. Analyze with AI and Structured content cost 1 credit plus their AI usage, at least 2. Podcast audio adds 0.25 credits per minute. Reading Results over MCP is free. New accounts start with free credits.

Questions

Short answers to what people ask most.

  • Does Owlie have an MCP server?

    Yes. Owlie hosts one. Create a personal access token in your account and give it to your MCP client. A read token lists and opens Results and Labels; a token that allows creating Operations can also set up new ones.

  • What does a webhook event contain?

    The Result (text or your structured fields), a link to it, the Operation, the Source, the item’s title, link and full text, and why it passed the filter. Send a test event from the Operation first.

  • Can I get YouTube transcripts or article text without a summary?

    Yes. Choose Extract only and each Result is the clean text: the article body, the video transcript, or the podcast episode transcribed.

  • Can I run the extractor myself?

    Yes. The extractor that reads web pages and videos is the open-source Owlie CLI. Run it locally when you only need text from a URL; use Owlie when you want it watched and delivered.

More use cases

Other ways to put Owlie to work.

Set this up in a few minutes.

Create an account, add a source, and choose the fields Owlie extracts from every new item.

Create accountSign in