Skip to main content

Google Vertex AI

Give Agent Designer agents access to your enterprise data context in DataHub — prototype data agents visually using the low-code builder in Vertex AI Agent Builder.

Note: Agent Designer is currently a preview feature.

Prerequisites​

  • A Google Cloud project with Vertex AI Agent Builder enabled
  • A DataHub instance with the MCP server enabled

Setup​

  1. Open Agent Designer and click Create agent.
  2. Set a name, instructions (e.g., "You are a data catalog assistant. Use DataHub tools to find datasets, schemas, and lineage."), and pick a model (e.g., Gemini 2.5 Flash).
  3. Click Add tools → MCP Server.
  4. Enter a display name (e.g., DataHub) and your MCP endpoint URL.
  5. Click Save — Agent Designer discovers the tools automatically.
  6. Use the Preview tab to test.

MCP Authentication Limitation​

The Agent Designer UI only supports MCP servers that do not require authentication. If your DataHub instance requires a bearer token, use the Get code button to export the agent, then add the Authorization header manually:

from google.adk.tools.mcp_tool import McpToolset
from google.adk.tools.mcp_tool.mcp_session_manager import StreamableHTTPConnectionParams

toolset = McpToolset(
    connection_params=StreamableHTTPConnectionParams(
        url="https://<tenant>.acryl.io/integrations/ai/mcp"
    ),
    headers={"Authorization": f"Bearer {YOUR_TOKEN}"},
)

See the Google ADK Integration for a complete working example.

Exporting to Code​

Click Get code to export your agent as Python, then continue development with the Google ADK or LangChain. This lets you prototype visually and transition to code for production.

Links: Agent Designer Docs · Google ADK Integration · Agent Context Kit