Discuss with AI Chat Feature Guide

There are two ways to get started: Download (easiest) and MCP (advanced).

For a single game, access it like so:

Option 1: Download (Easiest)

This works with any AI chatbot, free or paid, including ChatGPT, Claude, Gemini, and others.

How It Works

1. Open any processed game on PB Vision and click "Discuss with AI Chat"

2. Click the button to download your data file.

3. Open your favorite AI chatbot (e.g. ChatGPT)

4. Attach the downloaded data file to your chat interface and say “Help me analyze this game”.

The AI will ask which player you are, then dive into your game analysis. Start interacting with it immediately by saying: “I am [Name], analyze this game”

Multi-Game Analysis

Select multiple games or folders in your library and click “Discuss with AI” to see multiple games analyzed side by side

Tips for Copy/Paste

• Use the file attachment, not raw paste. The data is too large to paste as text for most games. Download the file and attach it to your chat instead.

• Longer games use more tokens. If your AI seems to struggle with a long game (20+ minutes), try asking focused questions about specific aspects like serves or third shots rather than a full breakdown.

• Any AI works. ChatGPT, Claude, Gemini, and others all support file attachments. Paid plans generally handle larger files and give better responses, but free tiers work for shorter games.

• Player names carry over. If players are tagged by name in your game, the AI will use those names automatically.

Video Walkthroughs

• How to use Discuss with AI (PB Vision walkthrough)

• John Kew's deep dive on the feature

Option 2: MCP (Advanced)

MCP (Model Context Protocol) is an open standard that lets AI assistants connect directly to external data sources. Instead of downloading files and attaching them, your AI pulls your PB Vision game data on demand. Just share a game link and start asking questions.

This option is best for users who are comfortable with technical setup and want a more seamless workflow.

Requirements

• An AI client that supports custom MCP servers (e.g., Claude Desktop with a Pro plan, Gemini Code Assist, Gemini CLI, or Cursor). Note: Web-based chat interfaces like ChatGPT and Gemini Advanced do not currently support custom MCP endpoints.

• A PB Vision account with uploaded and processed games.

Quick Start

MCP Server URL: https://share.pb.vision/mcp

No authentication or API key is needed. The server is an open endpoint you can connect to directly. The MCP server uses POST requests. If your client defaults to GET, make sure to configure it for POST.

Setup: Claude Desktop

Note: Claude Desktop connects to remote MCP servers through the Connectors menu.You'll need a paid Claude plan (Pro or Max).

1. Open Claude Desktop

2. Click the "+" button at the bottom of the chat input box

3. Select "Connectors"

4. Add a new custom connector with this URL:

   https://share.pb.vision/mcp

6. To verify the connection, start a new conversation, click the "+" button again, and select "Connectors" to confirm PB Vision is listed and active.

7. Then use it like so (make sure to Always allow it)

Setup: Gemini Advanced (CLI)

If you prefer working entirely in the terminal, the open-source Gemini CLI natively supports MCP using the exact same configuration engine as VS Code.

Note: Make sure you have Gemini Code CLI installed locally in your dev env. You will also need to make sure you are on a subscription with Google Gemini that allows for Code Assist (the free version does not).

1. Update your Gemini Settings File Open the same global configuration file used above:

• Mac/Linux: ~/.gemini/settings.json

• Windows: %USERPROFILE%\.gemini\settings.json

Add the remote server configuration:

JSON

{ "mcpServers": { "pb-vision": { "httpUrl": "https://share.pb.vision/mcp" } } } 

2. Activate and Verify

1. Open your terminal and launch the CLI by typing gemini.

2. Once the interactive prompt starts, type /mcp or /mcp list.

3. You should see pb-vision in the list with a green indicator showing it's ready .

4. You can now prompt Gemini directly from your command line to analyze your matches.

Setup: Gemini Code Assist (VS Code)

Gemini Code Assist connects to MCP servers through a local configuration file. This handshake allows the AI to pull your game data directly into your editor's sidebar.

Note: Make sure you have the Gemini Code Assist extension enabled in VS Code and you are signed in. You will also need to make sure you are on a subscription with Google Gemini that allows for Code Assist (the free version does not).

1. Locate your Gemini Settings File Open a text editor and find the settings.json file in your home directory:

• Mac/Linux: ~/.gemini/settings.json

• Windows: %USERPROFILE%\.gemini\settings.json

2. Add the PB Vision Configuration Paste the following JSON block into the file. If there is already a "mcpServers" section, just add the pb-vision entry to it.

{
  "mcpServers": {
    "pb-vision": {
      "httpUrl": "https://share.pb.vision/mcp"
    }
  }
}

Pro Tip: Ensure you have a folder or workspace open in VS Code. Gemini extensions often pause MCP connections in "empty" editor windows to save resources.

3. Activate and Verify

• Reload VS Code (Command Palette: Developer: Reload Window).

• Click the Gemini icon in the sidebar.

• Toggle Agent mode on (MCP tools are only accessible in Agent mode, not standard chat).

• Type /mcp in the chat to confirm pb-vision is listed as "Connected."

The interface should look like this:

Available MCP Tools

Once connected, your AI assistant gets access to five tools:

How to Use MCP

1. Find your video ID. This is the 12-character code in your game URL. For example, in https://pb.vision/video/ox31i0pijbzo/0/overview, the video ID is ox31i0pijbzo.

2. Start a conversation with your AI and share your video ID:

   • "Break down my play from game ox31i0pijbzo"

   • "Analyze my serve performance in game ox31i0pijbzo"

   • "Compare my stats across these two games: ox31i0pijbzo and 93vkk5bap5he"

3. The AI will use the MCP tools to pull your data and provide analysis. It will typically ask which player you are before diving in.

MCP Tips

• Start with the analysis guide. If the AI seems unsure how to interpret your data, ask it to run the get_analysis_guide tool first.

• Use download URLs for long games. For games over ~15 minutes, ask the AI to use get_insights_download_url instead of get_video_insights to keep token usage down.

• Multi-game analysis works. Share multiple video IDs in one conversation and ask the AI to compare your performance across games.

MCP Troubleshooting

"Connection failed" or no tools appearing Make sure your client is sending POST requests to the MCP URL, not GET. Restart your AI client after adding the server.

AI says it can't access the data Double-check that you're using the correct 12-character video ID from your game URL.

Response cuts off or seems incomplete Long games produce large data files. Ask the AI to use get_insights_download_url for a more token-efficient approach, or focus your questions on specific aspects of the game.

Need Help?

This feature is in beta and we're actively improving it. If you run into issues or have feedback:

• Email: support@pb.vision

• Discord: Join our community

• Feature requests: Vote and submit ideas