lilpacy/view-x-post

View X Post

Status Post Workflow

1. Fetch the post as Markdown:

npx curl.md "https://x.com/<handle>/status/<id>"

Use the URL the user provided. Do not install curl.md globally.

2. Verify the response is a real post page:

• Prefer the ## Post section as the canonical source.
• Cross-check title, description, author name, handle, and status URL.
• Treat login/signup links, "Don't miss what's happening", and other X chrome as noise.
• If the Markdown contains only login/signup/chrome and no post body, report that the post was not readable through curl.md.

3. Extract only what is supported by the Markdown:

• Author display name and handle
• Post body text
• Timestamp
• View/reply/repost/like/bookmark counts when present
• Attached media links such as /photo/1, /photo/2, etc.

Do not infer image contents from /photo/N links. Say only that media is attached unless another tool has actually inspected the media.

4. When answering, keep source uncertainty explicit:

• Say that the content was read via npx curl.md.
• If title/description and ## Post disagree, prefer ## Post and mention the mismatch.
• If metrics are shown as bare numbers without labels, avoid over-labeling them unless the surrounding order is clear.

X Article Workflow

Use chrome-devtools-mcp for https://x.com/<handle>/article/<id> URLs. npx curl.md usually returns only url and site for X Articles.

Use the headed Google Chrome profile named x-for-ai. Do not use headless mode for X Articles; measured attempts with headless did not preserve the logged-in X session reliably.

1. Open the article URL with chrome-devtools-mcp in the existing headed browser.

2. Verify the MCP browser is using the dedicated X profile:

• Open chrome://version.
• Confirm Profile Path is /Users/lilpacy/Library/Application Support/Google/Chrome/Profile 16.
• This is the Chrome profile named x-for-ai.
• If a different profile is attached, stop and ask the user to open or switch to the headed x-for-ai Chrome profile.

3. If X redirects to login:

• Bring the login tab to the front.
• Ask the user to log in in the browser window.
• Never ask for X credentials in chat and never type credentials yourself.
• Retry the article URL after the user confirms login.

4. Extract the article from the loaded page:

• Prefer take_snapshot first; X Articles expose title, author, headings, body text, metrics, and media links in the accessibility tree.
• If the snapshot is noisy, run evaluate_script and read document.querySelector('main').innerText.
• Keep media handling conservative: report image/video links or alt text shown by the page, but do not infer visual content without inspecting the media.

Example evaluate_script body:

() => {
  const main = document.querySelector("main");
  const text = main ? main.innerText : document.body.innerText;
  return {
    title: document.title,
    url: location.href,
    text,
    headings: Array.from(document.querySelectorAll("h1,h2"))
      .map((heading) => heading.innerText)
      .filter(Boolean),
  };
}

5. If the MCP browser is not using Profile 16, stop and report that the headed x-for-ai profile is not attached. Do not continue with another profile for authenticated X content.

Useful Commands

Fetch and save a temporary copy for inspection:

npx curl.md "https://x.com/<handle>/status/<id>" > /tmp/x-post.md

Search the fetched Markdown for the post area:

rg -n "^(title:|description:|## Post|@|[0-9:]+ [AP]M|[0-9]+\\.?[0-9]*[KM]?Views)" /tmp/x-post.md