VideoSDK Agents Docs MCP Server

The VideoSDK Agents Docs MCP Server is a standalone application that acts as a context provider for your favorite Model Context Protocol (MCP) compatible AI agents (like Claude Desktop, Cursor, or your own custom agent).

Instead of manually searching the VideoSDK documentation for answers, you can instruct your AI assistant to query the documentation natively. The MCP server ensures your assistant has instant access to up-to-date SDK references, implementation guides, and internal code examples to help you build your application faster.

Available Tools & Prompts

When connected, the server exposes the following capabilities to your AI assistant:

Tools

• get_details: Performs semantic search across the VideoSDK documentation allowing the AI to ask natural language questions (e.g., "How do I configure an AgentSession?").
• get_details_with_filter: Advanced semantic search that filters documents by category, platform, or type.
• get_document_by_id: Retrieves the full, un-truncated content and metadata of a specific VideoSDK documentation file.
• search_code: Scans the inner VideoSDK SDK source code to look up classes, methods, and functions (e.g., finding the exact arguments for a specific constructor).

Prompts

• cascading_ai_agent_prompt: Provides the AI agent with the context-rich prompt required to scaffold a Cascading AI Agent using VideoSDK.
• realtime_ai_agent_prompt: Provides the prompt needed to scaffold an ultra-low latency Realtime AI Agent.

---

Integration Guide

The VideoSDK Agents Docs MCP Server is hosted securely on the cloud and communicates via a streamable-http Connection. It does not require cloning a local repository or installing local dependencies.

You can connect your favorite MCP compatible client directly to our hosted endpoint.

Quick Setup by Client

Client	Connection Type	Configuration Method
Cursor	Direct streamable-http	GUI Settings
Claude Desktop	Bridge proxy (stdio)	JSON Config + Bridge Tool
VS Code (Roo Code/Cline)	Direct streamable-http	GUI Settings
VS Code (Continue)	Direct streamable-http	JSON Config
Antigravity	Direct streamable-http	GUI or JSON Config

Detailed Setup Instructions

Connecting via Cursor

To configure Cursor to use the VideoSDK Agents Docs MCP Server:

1. Open Cursor Settings and navigate to Features > MCP.
2. Click on + Add New MCP Server.
3. Provide a name (e.g., VideoSDK Docs).
4. Select the appropriate connection type for remote URLs (e.g., streamable-http).
5. Enter the VideoSDK MCP Server URL: https://mcp.videosdk.live/mcp.
6. Save and restart Cursor. The VideoSDK tools will now be available to Cursor's AI.

Connecting via Claude Desktop

Claude Desktop supports MCP servers through stdio connections. However, since the VideoSDK Agents Docs MCP Server is closed-source and only available remotely, you'll need to use a local proxy or bridge to connect:

Using a Local MCP Bridge

Since Claude Desktop requires stdio connections and the VideoSDK Agents Docs MCP Server is only available remotely, you can use an MCP bridge tool to proxy the remote server locally:

1. Install an MCP bridge tool like mcp-remote or mcp-proxy:

   npm install -g mcp-remote
   # or
   pip install mcp-bridge

2. Start the bridge pointing to the VideoSDK Agents Docs MCP Server:

   mcp-remote --url https://mcp.videosdk.live/mcp --port 3000

3. Open Claude Desktop.

4. Navigate to Settings > Connectors (or "Developer" in older versions).

5. Click Add Custom Connector (or "Add Connector").

6. Provide a name for the server (e.g., VideoSDK Docs).

7. For the command, enter:

   mcp-bridge --port 3000

8. Save the configuration and restart Claude Desktop. The VideoSDK tools will now appear as available attachments in your chats.

Note: If MCP bridge tools are not available or compatible with your setup, the VideoSDK Agents Docs MCP Server may not be accessible through Claude Desktop at this time. Consider using Cursor or VS Code extensions instead, which support direct remote streamable-http connections.

Connecting via VS Code

Since VS Code relies on extensions for AI features, here is how you can configure popular extensions like Roo Code (Cline) or Continue:

1. Roo Code (Cline):

   • Click the MCP Servers icon in the sidebar.
   • Click Add New Server and select the connection type for remote URLs.
   • Enter the name and the URL: https://mcp.videosdk.live/mcp.
   • Save to activate.

2. Continue.dev:

   • Open your config.json via the Continue extension settings.
   • Add the server URL under the mcpServers object as supported by their latest streamable protocol schema.

Connecting via Antigravity

To configure Antigravity to use the VideoSDK Agents Docs MCP Server, you have two options:

Option 1: GUI Configuration

1. Open Antigravity settings and navigate to MCP Configuration.
2. Click on Add MCP Server.
3. Provide a server name (e.g., VideoSDK Docs).
4. Select streamable-http as the connection type.
5. Enter the VideoSDK MCP Server URL: https://mcp.videosdk.live/mcp.
6. Save the configuration and restart Antigravity. The VideoSDK tools will now be available.

Option 2: JSON Configuration

1. Open ~/.gemini/antigravity/mcp_config.json
2. Add the following configuration:

{
    "mcpServers": {
        "videosdkAgentDocs": {
            "serverUrl": "https://mcp.videosdk.live/mcp"
        }
    }
}

JSON Configuration Snippets

For clients and IDE tools that accept direct JSON configuration editing, you can use the following configuration snippets:

General MCP Configuration

For tools like VS Code's cline_mcp_settings.json or other general mcpServers.json files:

{
  "mcpServers": {
    "videosdkAgentDocs": {
      "transport": "streamable-http",
      "url": "https://mcp.videosdk.live/mcp"
    }
  }
}

Claude Desktop Configuration

For Claude Desktop, which requires stdio-based MCP server connections, add the following to your claude_desktop_config.json file (typically located in ~/Library/Application Support/Claude/ on macOS):

{
  "mcpServers": {
    "videosdkAgentDocs": {
      "command": "mcp-bridge",
      "args": ["--url", "https://mcp.videosdk.live/mcp"],
      "env": {
        "NODE_ENV": "production"
      }
    }
  }
}

note

This assumes you have mcp-bridge installed globally via npm (npm install -g mcp-bridge). If using a different bridge tool, adjust the command and args accordingly. Make sure to restart Claude Desktop after updating the configuration.

Using Custom Clients

If you are building your own client, simply instruct your SDK to connect using the streamable-http URL: https://mcp.videosdk.live/mcp.

note

If the MCP server does not connect, try restarting your CLI or application. Some clients require a restart to properly initialize MCP server connections.

Example Usage

Once the MCP server is connected, you can use prompts like the following with your AI assistant:

"Claude, use the VideoSDK Agents Docs MCP Server to figure out how to add an OpenAI TTS plugin to my agent pipeline."

"Can you find the definition of the JobContext class using the search code tool, and write a script to initialize it?"

"Please load the realtime_ai_agent_prompt from the MCP Server and generate a basic voice agent."

The AI assistant will automatically coordinate with the MCP server to fetch the exact documents and code required to fulfill your prompt.