linkedin-weekly

name	linkedin-weekly
description	Use when collecting LinkedIn posts from followed targets and generating a weekly newsletter digest with AI summaries

LinkedIn Weekly Newsletter

Overview

Collect new LinkedIn posts from configured targets using Chrome DevTools MCP (real browser), summarize each, and generate a weekly HTML newsletter.

Prerequisites

• User must be logged into LinkedIn in their Chrome browser
• Chrome DevTools MCP must be connected

Workflow

Step 1: Get targets

Call list_targets to get all configured targets. Each target includes an activity_url field -- this is the URL to visit for that target's posts.

Step 2: Scrape each target via Chrome DevTools

For each target:

1. Navigate: Use navigate_page (Chrome DevTools MCP) to open the target's activity_url
2. Wait for content: Use wait_for to wait for post content to appear (e.g. text "activity" or a known page element)
3. Login wall check: Use take_snapshot and check for login wall indicators (text like "Sign in", "Join now", or no post content). If detected, stop and inform the user:

   "LinkedIn requires login. Please log into LinkedIn in your Chrome browser, then try again."

4. Scroll once: Use evaluate_script to scroll down once and load a few more posts:

   async () => {
     window.scrollBy(0, 1500);
     await new Promise(r => setTimeout(r, 2000));
   }
   

5. Extract posts via JS: Use evaluate_script to extract the top 3 posts directly from the DOM. This avoids large snapshots and returns compact JSON with stable IDs:

   () => {
     const posts = document.querySelectorAll(
       '[data-urn*="urn:li:activity"], .feed-shared-update-v2, .occludable-update'
     );
     return Array.from(posts).map(el => {
       const urn = el.getAttribute('data-urn') || '';
       const idMatch = urn.match(/urn:li:activity:\d+/);
       const authorEl = el.querySelector(
         '.update-components-actor__title span[aria-hidden="true"], ' +
         '.update-components-actor__name span[aria-hidden="true"], ' +
         '.feed-shared-actor__name span[aria-hidden="true"]'
       );
       const textEl = el.querySelector(
         '.feed-shared-update-v2__description, ' +
         '.update-components-text, .feed-shared-text'
       );
       const linkEl = el.querySelector('a[href*="feed/update"]');
       const timeEl = el.querySelector('time');
       const subDesc = el.querySelector('.update-components-actor__sub-description');
       const imgs = el.querySelectorAll(
         '.feed-shared-image__image, .update-components-image img'
       );
       const linkedinId = idMatch ? idMatch[0] : null;
       let postedAt = '';
       if (timeEl) {
         postedAt = timeEl.getAttribute('datetime') || timeEl.innerText.trim();
       } else if (subDesc) {
         postedAt = subDesc.innerText.trim().split('•')[0].trim();
       }
       return {
         linkedin_id: linkedinId,
         author: authorEl ? authorEl.innerText.trim() : 'Unknown',
         text: textEl ? textEl.innerText.trim().slice(0, 500) : '',
         url: linkEl ? linkEl.href : (linkedinId ? 'https://www.linkedin.com/feed/update/' + linkedinId : ''),
         media_urls: Array.from(imgs).map(i => i.src).filter(Boolean),
         posted_at: postedAt,
       };
     }).filter(p => p.linkedin_id).slice(0, 3);
   }
   

6. Fallback: If the JS extraction returns an empty array (e.g. LinkedIn changed selectors), fall back to take_snapshot of the current visible area only (no additional scrolling) and extract posts from the snapshot text. Use snapshot-extracted data as best-effort -- IDs may not be stable.
7. Store posts: Call store_posts(target_url=<target's base url>, posts=<extracted posts list>)

Step 3: Generate newsletter

1. Call get_new_posts(since_days=7) to retrieve all posts from the past week
2. For each post, generate a one-sentence English summary
3. If original post text is non-English, also generate an English translation
4. Group posts by target_name, sort by posted_at (newest first)
5. Call generate_newsletter passing the list of posts, each with added summary field (and translation field if applicable)
6. Call open_newsletter with the returned file path
7. Report to user: total targets checked, new posts found, newsletter file path

Summary Guidelines

• Always write summaries in English regardless of original language
• One sentence per post capturing the core point
• If post contains links or media, mention briefly
• Keep original post text in its original language
• For non-English posts, add a translation field with full English translation