Post to X via Browser

Publish a post or thread directly to X using chrome browser automation.

Input

The user provides one of:

• Post content directly as the argument
• "last" to use the output from a previous /compose or /thread invocation in this session
• A thread as numbered tweets or clearly separated tweets
• Media to attach — a file path (image, GIF, video), a screenshot taken this session, or a request to screenshot a URL first via /media

If the content includes a link reply (a separate reply containing a URL), handle that as a follow-up reply after the main post.

Pre-Flight

Before doing anything:

1. Parse the input to determine: single post vs thread vs post + link reply vs post + media
2. Validate character counts (each tweet ≤ 280 characters)
3. If any tweet exceeds 280 characters, stop and ask the user to revise — do NOT truncate
4. If media is referenced, determine the media type and upload path (see Step 5)

Process

Step 1 — Get Browser Context

mcp__claude-in-chrome__tabs_context_mcp(createIfEmpty: true)

Check if there's an existing tab on x.com. If yes, use it. If not, create a new tab.

Step 2 — Navigate to X

mcp__claude-in-chrome__navigate(url: "https://x.com", tabId: <tab>)

Wait for the page to load. Take a screenshot to verify you're on X and logged in.

mcp__claude-in-chrome__computer(action: "screenshot", tabId: <tab>)

If not logged in: Stop and tell the user they need to log in manually. Do NOT attempt to handle authentication.

Step 3 — Open Compose

Navigate directly to the compose URL (most reliable):

mcp__claude-in-chrome__navigate(url: "https://x.com/compose/post", tabId: <tab>)

Wait for the compose dialog to fully load:

mcp__claude-in-chrome__computer(action: "wait", duration: 2, tabId: <tab>)

Step 4a — Single Post

1. Find the text input area in the compose dialog:

mcp__claude-in-chrome__find(query: "tweet text input area", tabId: <tab>)

1. Click on it to focus:

mcp__claude-in-chrome__computer(action: "left_click", ref: "<ref_id>", tabId: <tab>)

1. Type the post content:

mcp__claude-in-chrome__computer(action: "type", text: "<post_content>", tabId: <tab>)

1. If media needs to be attached, do Step 5 now (before the review loop).

1. Enter the Review Loop (see Review Loop section below).

Step 4b — Thread

1. Open compose and type the first tweet (same as 4a steps 1-3).

1. To add the next tweet, find and click the "Add another post" button (the + icon):

mcp__claude-in-chrome__find(query: "add another tweet button or plus button to add to thread", tabId: <tab>)
mcp__claude-in-chrome__computer(action: "left_click", ref: "<ref_id>", tabId: <tab>)

1. Type the next tweet's content in the new text area:

mcp__claude-in-chrome__find(query: "empty tweet text input", tabId: <tab>)
mcp__claude-in-chrome__computer(action: "left_click", ref: "<ref_id>", tabId: <tab>)
mcp__claude-in-chrome__computer(action: "type", text: "<tweet_content>", tabId: <tab>)

1. Repeat steps 2-3 for each subsequent tweet in the thread.

1. If media needs to be attached to any tweet, do Step 5 for that tweet.

1. Enter the Review Loop (see Review Loop section below).

Step 4c — Link Reply (after posting)

If the original content included a link that should go in a reply:

1. After the main post is published, wait for it to appear:

mcp__claude-in-chrome__computer(action: "wait", duration: 3, tabId: <tab>)

1. Find the just-published post and click the reply icon:

mcp__claude-in-chrome__find(query: "reply button on the latest post", tabId: <tab>)
mcp__claude-in-chrome__computer(action: "left_click", ref: "<ref_id>", tabId: <tab>)

1. Type the link reply content:

mcp__claude-in-chrome__computer(action: "type", text: "<link_reply_content>", tabId: <tab>)

1. Take a screenshot and enter the Review Loop for the reply too.

Review Loop

This is the critical confirmation flow. Every post and reply goes through this loop before publishing.

1. Screenshot the current state

mcp__claude-in-chrome__computer(action: "screenshot", tabId: <tab>)

For threads, scroll through to capture all tweets if needed.

2. Present the review to the user

Use AskUserQuestion to present the post and collect feedback:

AskUserQuestion:
  "Here's your [post/thread/reply] ready to publish:

  [Show the full text of what will be posted]

  ✅ 'post' — publish now
  ✏️ Reply with changes — I'll revise and show you again
  ❌ 'cancel' — abort without posting"

3. Handle the response

If the user says "post", "yes", "go", "send it", or similar affirmative: → Proceed to click the Post button:

mcp__claude-in-chrome__find(query: "post button or tweet button", tabId: <tab>)
mcp__claude-in-chrome__computer(action: "left_click", ref: "<ref_id>", tabId: <tab>)

Wait for publish:

mcp__claude-in-chrome__computer(action: "wait", duration: 3, tabId: <tab>)

Take confirmation screenshot:

mcp__claude-in-chrome__computer(action: "screenshot", tabId: <tab>)

If the user requests changes (e.g., "make the hook stronger", "remove the last line", "add a question at the end"): → Apply the requested changes:

1. Select all text in the compose box (triple-click or Cmd+A)
2. Delete it
3. Type the revised content
4. Loop back to step 1 — screenshot and present the revision for review

The loop continues until the user approves or cancels. There is no limit on revision rounds.

If the user says "cancel", "abort", "nevermind": → Close the compose dialog (press Escape or navigate away) → Confirm: "Post cancelled. The draft is still available if you want to revise it."

Step 5 — Media Attachment

Media attachment happens BEFORE the review loop, so the user sees the full post with media in the screenshot.

Path A: Screenshot from this session

If the media is a screenshot captured during this Claude Code session (e.g., from /media taking a screenshot of a webpage):

1. Find the media upload button in compose:

mcp__claude-in-chrome__find(query: "add photo or media button in compose", tabId: <tab>)
mcp__claude-in-chrome__computer(action: "left_click", ref: "<ref_id>", tabId: <tab>)

1. Find the file input element:

mcp__claude-in-chrome__find(query: "file input for media upload", tabId: <tab>)

1. Upload via the ref:

mcp__claude-in-chrome__upload_image(imageId: "<screenshot_id>", ref: "<file_input_ref>", tabId: <tab>)

1. Wait for the upload to process:

mcp__claude-in-chrome__computer(action: "wait", duration: 3, tabId: <tab>)

Path B: Local file (Remotion video, image, GIF)

For local files on disk (e.g., a Remotion-rendered .mp4 at marketing/out/video.mp4):

The chrome upload_image tool cannot directly access local filesystem files. Use this approach:

1. Find the media upload button and click it:

mcp__claude-in-chrome__find(query: "add photo or media button in compose", tabId: <tab>)
mcp__claude-in-chrome__computer(action: "left_click", ref: "<ref_id>", tabId: <tab>)

1. This opens a native file picker dialog. Tell the user:

AskUserQuestion:
  "I've opened the media upload dialog. Please select your file:

  📁 [file path, e.g., marketing/out/video.mp4]

  Navigate to this file in the picker and click Open.
  Let me know once the file is selected and uploading."

1. After the user confirms the file is selected, wait for processing:

mcp__claude-in-chrome__computer(action: "wait", duration: 5, tabId: <tab>)

1. Take a screenshot to verify the media attached correctly:

mcp__claude-in-chrome__computer(action: "screenshot", tabId: <tab>)

1. If the media looks wrong, ask the user if they want to remove it and try again.

Path C: URL screenshot (inline)

If the user wants to screenshot a webpage to attach:

1. Navigate to the URL in a separate tab:

mcp__claude-in-chrome__tabs_create_mcp()
mcp__claude-in-chrome__navigate(url: "<url>", tabId: <new_tab>)
mcp__claude-in-chrome__computer(action: "wait", duration: 3, tabId: <new_tab>)

1. Take a screenshot (this creates an imageId):

mcp__claude-in-chrome__computer(action: "screenshot", tabId: <new_tab>)

1. Switch back to the X compose tab and use Path A to upload the screenshot.

Media Compatibility

Post-Publish

After successful posting:

1. Take a confirmation screenshot showing the published post
2. If there's a link reply, proceed to Step 4c
3. Remind the user:

"Posted! 🎯

Remember: reply to every comment in the first hour — that's the 75–150× conversation velocity multiplier.

Run /engage on this post in 30-60 minutes to manage replies."

Dry-Run Mode

If the user says "dry run", "preview only", or "don't actually post":

• Complete the entire flow up through the Review Loop (navigate, compose, type, screenshot)
• Stop before clicking Post. Show the screenshot and say: "Dry run complete — here's what would be posted. Nothing was published."
• This lets users test the full pipeline without risk.

Rate Limiting

• Wait 3-5 seconds between any two posting actions (post → link reply, or thread → next action)
• Never post more than 5 times in a single /post invocation (threads count as 1)
• If X shows an error or rate limit message, screenshot it, inform the user, and stop. Do not retry.

Critical Rules

1. Every post goes through the Review Loop. No exceptions.
2. The user can request unlimited revisions. Loop until they approve or cancel.
3. NEVER handle authentication. If not logged in, tell the user to log in manually.
4. NEVER exceed 280 characters per tweet. Validate before typing.
5. If anything goes wrong (element not found, page not loading, unexpected state), take a screenshot, describe what you see, and ask the user how to proceed. Do not retry blindly.
6. For local file uploads, always tell the user the exact file path so they can find it in the picker.
7. X may detect automation. Browser automation leaves different fingerprints than manual use. Always review screenshots before confirming. If X presents a CAPTCHA or verification challenge, stop and inform the user.

Troubleshooting

• Can't find compose button: Navigate directly to x.com/compose/post
• Can't find post button: Take a screenshot, look for the button visually, use coordinates
• Thread + button not found: Look for icon buttons near the compose area, may be labeled differently
• Page loads slowly: Use wait with 2-3 seconds between navigation and interaction
• Content has special characters: Type content carefully; if emojis fail, try javascript_tool to set the value directly via DOM
• Media upload stuck: Take a screenshot to check status, wait up to 10 seconds for video processing
• File picker won't close: The user may need to manually dismiss it; ask them
• Video too large: X accepts up to 512MB / 2:20 duration — suggest trimming with ffmpeg if needed