threads-search

name	threads-search
description	Threads (threads.com) search & data extraction — cari post, extract comments/replies, scroll pagination. Browser-based via Patchright MCP.
triggers	["cari di threads","search threads","threads search","research threads","cari data threads","threads-research","cari post threads","cari utas"]
tools	["mcp_patchright_browser_navigate","mcp_patchright_browser_click","mcp_patchright_browser_snapshot","mcp_patchright_browser_run_code","mcp_patchright_browser_wait_for","mcp_patchright_browser_scroll","mcp_patchright_browser_tabs"]

Threads Search & Data Extraction

Search Threads (threads.com), extract posts with full reply/comment threads via Patchright MCP.

Prerequisites

• Patchright daemon running (port 9877)
• Profile logged in to Threads (shared with Instagram)
• Default profile: rickicode

Search URL Patterns

Filter	URL
Top (default)	https://www.threads.com/search?q={query}
Recent	https://www.threads.com/search?q={query}&filter=recent
Profiles	https://www.threads.com/search?q={query}&filter=profiles
Author posts	https://www.threads.com/search?from_author={username}
Tag search	https://www.threads.com/search?q={tag}&serp_type=tags&tag_id={id}

Extract Search Results — VERIFIED WORKING

async (page) => {
  await page.waitForTimeout(5000);
  return await page.evaluate(() => {
    const results = [];
    const seen = new Set();
    const links = document.querySelectorAll('a[href*="/post/"]');

    for (const link of links) {
      const postUrl = link.href;
      if (seen.has(postUrl)) continue;
      seen.add(postUrl);

      // Walk up to find the containing post block
      const container = link.closest('[data-pressable-container]') || link.closest('div[class*="post"]') || link.parentElement?.parentElement?.parentElement;
      if (!container) continue;

      const text = container.innerText || '';
      const timeEl = container.querySelector('time');
      const authorLink = container.querySelector('a[href^="/@"]');

      results.push({
        authorHandle: authorLink?.textContent?.trim() || '',
        authorUrl: authorLink?.href || '',
        text: text.split('\n').filter(l => l.length > 10).slice(0, 3).join(' | ').substring(0, 500),
        postUrl,
        timestamp: timeEl?.getAttribute('datetime') || '',
        timeDisplay: timeEl?.textContent?.trim() || '',
      });
    }
    return results;
  });
}

Extract Post with Replies (Detail Page) — VERIFIED WORKING

Navigate to post URL, then extract:

async (page, postUrl) => {
  await page.goto(postUrl);
  await page.waitForTimeout(5000);

  // Scroll to load more replies
  for (let i = 0; i < 3; i++) {
    await page.evaluate(() => window.scrollBy(0, 800));
    await page.waitForTimeout(2000);
  }

  return await page.evaluate(() => {
    const results = { post: null, replies: [] };
    const items = document.querySelectorAll('[data-pressable-container="true"]');

    for (const item of items) {
      const links = item.querySelectorAll('a');
      const authorLink = [...links].find(a => a.href?.includes('/@') && !a.href?.includes('/post/'));
      const timeEl = item.querySelector('time');
      const text = item.innerText || '';

      // Parse engagement counts from buttons
      const btns = item.querySelectorAll('button');
      let likeCount = '', replyCount = '', repostCount = '';
      for (const btn of btns) {
        const label = btn.getAttribute('aria-label') || btn.textContent || '';
        if (label.startsWith('Like') && /\d/.test(label)) likeCount = label.replace('Like', '').trim();
        if (label.startsWith('Reply') && /\d/.test(label)) replyCount = label.replace('Reply', '').trim();
        if (label.startsWith('Repost') && /\d/.test(label)) repostCount = label.replace('Repost', '').trim();
      }

      const entry = {
        authorHandle: authorLink?.textContent?.trim() || '',
        authorUrl: authorLink?.href || '',
        text: text.split('\n').filter(l => l.length > 10).slice(0, 3).join(' | ').substring(0, 500),
        postUrl: [...links].find(a => a.href?.includes('/post/'))?.href || '',
        timestamp: timeEl?.getAttribute('datetime') || '',
        timeDisplay: timeEl?.textContent?.trim() || '',
        likes: likeCount,
        replies: replyCount,
        reposts: repostCount,
      };

      if (!results.post) {
        results.post = entry;
      } else if (entry.text.length > 5) {
        results.replies.push(entry);
      }
    }

    return results;
  });
}

Extract Profile Posts — VERIFIED WORKING

async (page, username) => {
  await page.goto(`https://www.threads.com/@${username}`);
  await page.waitForTimeout(5000);

  for (let i = 0; i < 3; i++) {
    await page.evaluate(() => window.scrollBy(0, 1000));
    await page.waitForTimeout(2000);
  }

  return await page.evaluate(() => {
    const followerText = document.body.innerText.match(/(\d[\d,.]*[KMB]?)\s*followers/i);

    const postLinks = document.querySelectorAll('a[href*="/post/"]');
    const seen = new Set();
    const posts = [];

    for (const link of postLinks) {
      if (seen.has(link.href)) continue;
      seen.add(link.href);

      const container = link.closest('[data-pressable-container]') || link.parentElement?.parentElement;
      const text = container?.innerText || '';
      const timeEl = container?.querySelector('time');

      posts.push({
        text: text.split('\n').filter(l => l.length > 10).slice(0, 3).join(' | ').substring(0, 300),
        postUrl: link.href,
        timestamp: timeEl?.getAttribute('datetime') || '',
        timeDisplay: timeEl?.textContent?.trim() || '',
      });
    }

    return {
      followers: followerText?.[1] || '',
      posts
    };
  });
}

Scroll Pagination — VERIFIED WORKING

async (page, maxScrolls = 5) => {
  let allPosts = [];
  const seen = new Set();

  for (let i = 0; i < maxScrolls; i++) {
    const posts = await page.evaluate(() => {
      const links = document.querySelectorAll('a[href*="/post/"]');
      return [...links].map(l => ({
        url: l.href,
        text: (l.closest('[data-pressable-container]')?.innerText || '').substring(0, 100),
      }));
    });

    for (const p of posts) {
      if (p.url && !seen.has(p.url)) {
        seen.add(p.url);
        allPosts.push(p);
      }
    }

    await page.evaluate(() => window.scrollBy(0, 1500));
    await page.waitForTimeout(2000);
  }

  return { total: allPosts.length, posts: allPosts };
}

Pitfalls

1. Column layout: Threads uses columns, not full pages. Content loads in active column.
2. English UI: All labels in English (Like, Reply, Repost, Share, Follow, Post).
3. Ref staleness: After navigation, always get fresh snapshot.
4. Post URL format: /@{user}/post/{shortcode}.
5. Engagement counts: In button text — "Like 36", "Reply 6", "Repost 4".
6. "Show more": Long posts have expand button — click to get full text.
7. DM URL: /messages/t/{thread_id}/ — different from Instagram.
8. Shared auth: Threads shares login with Instagram.
9. Rate limiting: Add waitForTimeout(3000) between navigations.
10. No views count on search: Only visible on post detail page.
11. Post detail selector: [data-pressable-container="true"] — first item is main post, rest are replies.
12. Media URL: Images from cdninstagram domain.
13. Translate button: Non-English posts show "Translate" button.
14. "No replies yet": New posts show "No replies yet" text — replies array will be empty.
15. View count: Post detail shows "1 view", "2.5K views" in column header area.
16. Fediverse indicator: Some posts show "Post will be shared to Fediverse" icon.
17. Tab loading: After navigate, Threads takes 3-6s to load. Use waitForTimeout(5000).
18. Post link dedup: Search results may include /post/xxx/media URLs — deduplicate by base post URL.
19. File upload path: Patchright only allows files from /workspaces/patchright-browser/. Copy files there before uploading.
20. Duplicate image upload: fileChooser.setFiles() sometimes uploads 2 copies. Check Remove button count after upload.
21. Character limit ~500: Posts over ~500 chars show negative counter and won't publish.
22. MCP client from Python: Use mcp.client.streamable_http.streamablehttp_client + ClientSession — NOT raw HTTP POST. See patchright-browser skill for full pattern.
23. Image upload duplication: When uploading images to Threads composer via Patchright, the file chooser can trigger twice, creating 2 copies. After upload, check for duplicate "Remove" buttons — if 2 exist, click one to remove the duplicate before posting.