ADK/LangGraph MCP A2A YouTube Summarizer

This project demonstrates inter-agent communication using the A2A protocol between an agent built with Google's Agent Development Kit (ADK) and another built with LangGraph. It uses real MCP tool endpoints and Google Cloud Secret Manager for API key handling.

• LangGraph YouTube Agent: Finds YouTube video IDs based on channel/date or playlist ID using specific MCP tool URLs. Runs an A2A server.
• ADK Summary Agent: Coordinates the summarization process. It receives a user request, delegates video finding to the YouTube Agent via A2A, receives the video IDs, calls specific MCP tool URLs to summarize each video, combines the summaries using another MCP tool URL, and returns the final result (streaming updates). Runs an A2A server.

Prerequisites

• Python 3.11+
• uv (recommended Python package manager)
• Google Cloud Project with Secret Manager API enabled.
• A secret stored in Google Cloud Secret Manager containing your Google AI (Gemini) API Key.
• URLs for four running MCP servers (or fewer if tools are hosted together):
  • get_playlist_videos tool endpoint.
  • get_youtube_videos_for_channel_date tool endpoint.
  • get_youtube_video_summary tool endpoint.
  • generate_final_summary tool endpoint.
• (Optional) Service account JSON files if your MCP servers require authentication.
• a2a-samples library installed or common directory copied from the google-a2a repository.

Setup

1. Clone the repository (or run the creation script):

   cd adk-langgraph-a2a-youtube-summarizer

2. Install Dependencies:

   uv venv
   source .venv/bin/activate
   # Ensure a2a-samples is installed or common dir setup in pyproject.toml
   uv pip install -e . # Install root package and workspace members

3. Configure Environment Variables:

   • Fill in the placeholders in langgraph_youtube_agent/.env. Remove GOOGLE_API_KEY. Add SECRET_PROJECT_ID and GOOGLE_API_KEY_SECRET_ID. Add MCP_URL_GET_PLAYLIST and MCP_URL_GET_CHANNEL.
   • Fill in the placeholders in adk_summary_agent/.env. Remove GOOGLE_API_KEY. Add SECRET_PROJECT_ID and GOOGLE_API_KEY_SECRET_ID. Add MCP_URL_SUMMARIZE and MCP_URL_COMBINE. Ensure YOUTUBE_AGENT_A2A_URL is correct. Replace placeholders.

4. Authentication: Ensure your environment is authenticated to Google Cloud with permissions to access the specified secret in Secret Manager (e.g., run gcloud auth application-default login).

Running

1. Start the MCP Servers: Ensure your MCP servers are running and accessible at the URLs specified in the .env files.

2. Start the LangGraph YouTube Agent A2A Server: (Terminal 1)

   # From the root directory, activate venv
   uv run langgraph_youtube_agent --port 10003

3. Start the ADK Summary Agent A2A Server: (Terminal 2)

   # From the root directory, activate venv
   uv run adk_summary_agent --port 10004

4. Interact using an A2A Client: (Terminal 3) Use an A2A client pointed at the ADK Summary Agent's URL (http://localhost:10004).

   # Example using the sample CLI
   # cd <path_to_google_a2a>/samples/python/hosts/cli
   # Activate venv if needed
   # uv run . --agent http://localhost:10004 --stream

   Enter prompts like:

   • Summarize videos from channel <channel_id> on <YYYY-MM-DD>
   • Give me a summary for playlist <playlist_id>

   Observe logs and streaming output.