Skip to content
QuickFlo Docs

Webhook → AI URL Summary

A webhook URL that takes { "url": "https://..." }, fetches the page, runs an LLM Call in structured output mode to extract a typed summary, and returns clean JSON. AI-powered Cliffs Notes for any URL — drop it into a Slack bot, a browser extension, or a Notion automation, and you get back structured data you can pipe into anything.

Workflow JSON

Section titled “Workflow JSON”

Before pasting, swap the "connection": "my-anthropic" placeholder for the name of your actual Anthropic (or OpenAI / Google) connection.

webhook-ai-url-summary.json
{
  "name": "URL Summary API",
  "initial": {
    "url": "https://example.com"
  },
  "steps": [
    {
      "stepId": "fetch-page",
      "stepType": "core.http",
      "input": {
        "url": "{{ initial.url }}",
        "method": "GET",
        "retry": {
          "limit": 3
        }
      }
    },
    {
      "stepId": "summarize",
      "stepType": "ai.llm-call",
      "input": {
        "provider": {
          "provider": "anthropic",
          "connection": "my-anthropic",
          "model": "claude-sonnet-4-5-20250929"
        },
        "mode": {
          "mode": "structured_output",
          "system_prompt": "You are a concise content analyst. Extract a clean structured summary from the HTML or text the user gives you. Be terse and accurate.",
          "user_prompt": "Summarize the following content from {{ initial.url }}:\n\n{{ fetch-page.text }}",
          "output_schema": {
            "title": "string: The page's title",
            "summary": "string: A 2-3 sentence summary",
            "topics": "string[]: A short list of topical tags",
            "sentiment": "string: One of 'positive', 'neutral', 'critical', 'mixed'",
            "wordCount": "number: Approximate word count of the original page"
          }
        }
      }
    },
    {
      "stepId": "respond",
      "stepType": "core.return",
      "input": {
        "webhookResponse": {
          "statusCode": 200,
          "body": "{{ summarize.object }}"
        }
      }
    }
  ]
}

Setup

Section titled “Setup”

Connection needed: an Anthropic, OpenAI, or Google API key connection. Anthropic Claude is the default in the recipe — swap the connection name and provider if you use OpenAI or Google.

Trigger: add a Webhook trigger to the workflow with:

FieldValue
Namesummarize-url
MethodPOST
AuthenticationRecommended on. Generate a webhook secret and pass it as a Bearer token.

How it works

Section titled “How it works”

Three steps:

  1. Fetch the page — HTTP GET against the URL the caller passed in.
  2. Summarize with structured output — LLM Call configured to return JSON matching a schema. Instead of asking the model for free text and parsing it, you define an output schema and the model is constrained to return JSON matching it. No regex, no JSON-extraction headaches, no hallucinated keys.
  3. Return the summary — webhook response with the structured object.

Sample response:

{
  "title": "Why writing a SaaS in Rust is harder than you think",
  "summary": "A blog post arguing that Rust's strict type system creates friction for the kind of rapid iteration most B2B SaaS startups need...",
  "topics": ["rust", "saas", "type systems", "developer productivity"],
  "sentiment": "critical",
  "wordCount": 1840
}

Test it

Section titled “Test it”
Terminal window
curl -X POST https://run.quickflo.app/w/@your-org/summarize-url \
  -H "Authorization: Bearer YOUR_WEBHOOK_SECRET" \
  -H "Content-Type: application/json" \
  -d '{"url": "https://en.wikipedia.org/wiki/Workflow_engine"}'

What to customize first

Section titled “What to customize first”
  • Strip HTML before sending to the model. Right now fetch-page.text includes raw HTML. Add a set-variable step that uses Liquid’s strip_html filter to clean it up first — saves tokens and improves quality.
  • Cap input size. Long pages will blow past the model’s context window. Truncate fetch-page.text to the first ~30,000 characters with {{ fetch-page.text | substring: 0, 30000 }} before passing it in.
  • Add a knowledge base for company-specific context. If you’re summarizing internal docs, attach a knowledge base to the LLM Call so the model has background context.
  • Switch to AI Agent for multi-step extraction. If you want the model to also fetch a related URL or query a CRM as part of summarizing, swap the LLM Call for an AI Agent and define those capabilities as tools.
Section titled “Related recipes”