reagent-systems/mcp-x-posting

X Post Generator MCP Server

A standalone Python-based Model Context Protocol (MCP) server that integrates with Cursor to automatically generate draft X (Twitter) posts based on your coding progress. It captures screenshots of your browser preview, analyzes recent git changes, and uses a local AI model in LM Studio to craft engaging posts. Drafts are saved to a local CSV for easy review and editing.

This keeps your workflow private and local—no cloud services or extensions required.

Features

• Screenshot Capture: Automatically opens your app preview URL in the browser and takes a screenshot using pyautogui.
• Git Change Analysis: Pulls recent code diffs to understand features you're working on.
• Local AI Generation: Leverages LM Studio (with a vision model) to create concise, hashtag-filled X posts under 280 characters.
• CSV Storage: Saves drafts with timestamps, summaries, posts, and screenshot paths for spreadsheet import.
• MCP Integration: Seamless invocation from Cursor's chat or Composer (e.g., @x-post-generator create_x_post_draft).
• Cross-Platform: Works on Windows, Mac, and Linux.

Prerequisites

• Cursor: Download from cursor.com. No extensions needed—MCP is built-in.
• LM Studio: Download from lmstudio.ai. Load a vision-capable model like llama-3.2-vision-instruct and start the local server (default: http://localhost:1234).
• Python 3.10+: With pip.
• Git: For change detection (optional; falls back gracefully if not present).
• Browser: Default system browser for preview screenshots (assumes your dev server runs at http://localhost:3000—configurable).

Installation

1. Clone or Save the Script:

   • Save mcp_server.py from the provided code.

2. Set Up Virtual Environment:

   python -m venv mcp-env
   # Activate:
   # Linux/Mac: source mcp-env/bin/activate
   # Windows: mcp-env\Scripts\activate
   

3. Install Dependencies:

   pip install mcp[cli] openai pyautogui pillow gitpython pandas requests pytest-asyncio
   

   • Note: pyautogui may prompt for permissions on first run (e.g., screen recording on Mac).

4. Configure MCP in Cursor:

   • Open Cursor Settings (Cmd/Ctrl + ,) > Search "MCP".
   • Add this to your settings JSON (under "mcpServers"):

     {
       "mcpServers": {
         "x-post-generator": {
           "command": "python",
           "args": ["/absolute/path/to/mcp_server.py"],
           "env": {}
         }
       }
     }
     

   • Restart Cursor.

Configuration

Edit the top of mcp_server.py:

• LM_STUDIO_URL: Your LM Studio server (default: http://localhost:1234/v1).
• MODEL_NAME: Loaded model (e.g., llama-3.2-vision-instruct).
• BROWSER_URL: Your app preview (default: http://localhost:3000).
• DRAFTS_FILE: Output CSV path (default: ./x_post_drafts.csv).

Usage

1. Start the MCP Server:

   python mcp_server.py
   

   • It will check LM Studio connectivity and run in stdio mode for Cursor.

2. In Cursor:

   • Open Chat or Composer.
   • Prompt: @x-post-generator create_x_post_draft("Implemented user login feature")
     • This triggers the full workflow: screenshot → analysis → generation → save.
   • Or use sub-tools: @x-post-generator get_recent_changes for diffs only.

3. Review Drafts:

   • Open ./x_post_drafts.csv in Excel, Google Sheets, etc.
   • Columns: Timestamp, Feature Summary, Draft Post, Screenshot Path.
   • Example Draft: "🚀 Just nailed user auth in my app! Seamless login with OAuth. Check the clean UI below. #BuildInPublic #IndieDev [Screenshot desc]"

4. Workflow Example:

   • Code → Save/Commit in Cursor.
   • Prompt for draft.
   • Edit/post manually on X.

Available Tools

From Cursor, you can call:

• create_x_post_draft(feature_desc): Full pipeline (default: "latest work").
• get_recent_changes(): Git diff summary.
• take_screenshot(): Capture and return path/base64.
• generate_post(feature_summary, screenshot_path, screenshot_b64): AI post creation.
• save_draft(feature_summary, draft_post, screenshot_path): Append to CSV.

Troubleshooting

• LM Studio Not Connecting: Ensure server is running and model loaded. Test with curl http://localhost:1234/v1/models.
• Screenshot Fails: Grant pyautogui permissions. Increase asyncio.sleep(3) for slower loads. For headless: Swap to playwright (pip install; requires browser install).
• No Git Repo: Script handles it—provide manual feature_desc in prompts.
• MCP Not Found in Cursor: Double-check settings JSON path and restart.
• Long Prompts: Truncates diffs to 2000 chars; adjust in code.
• Testing: Add pytest for unit tests (e.g., mock API calls).

Enhancements

• Auto-Trigger: Add a Cursor task in .vscode/tasks.json to run on save.
• Google Sheets: Integrate gspread for cloud sync.
• Direct Posting: Extend with X API (add tweepy; requires keys—not included for privacy).

License

MIT License—feel free to fork and improve!

---

Built with ❤️ for indie devs sharing their journey. Questions? Open an issue or ping on X at https://www.x.com/reagent-systems/.