Demandbase MCP in Gemini Setup Guide

This guide explains how to connect the Demandbase MCP server to a custom ADK agent, deploy that agent to Vertex AI Agent Engine, and register it in Gemini Enterprise so Workspace users can access Demandbase tools from Gemini.

Architecture

• ADK agent: a Python LlmAgent built with google-adk
• Demandbase MCP server: remote Streamable HTTP endpoint exposing Demandbase intelligence tools
• Vertex AI Agent Engine: managed runtime for the deployed agent
• Gemini Enterprise: the surface where Workspace users interact with the agent

1. Prerequisites

Before you begin, make sure you have:

• A Google Cloud project with Vertex AI enabled
• Permission to create and deploy Vertex AI Agent Engine resources
• A supported Python version for ADK
• Access to the Demandbase MCP server credentials from your Demandbase admin
• Google Workspace admin access if you plan to register the agent in Gemini Enterprise

2. Set Up the Environment

1. Authenticate to Google Cloud. (Instructions)
2. Install the ADK Python package and create a virtual environment.
   1. ADK Python Github
   2. ADK open source docs
   3. ADK Quickstart (Vertex AI)
3. Verify that your Google Cloud project and billing settings are ready for Vertex AI.

Example setup:

gcloud auth login
gcloud config set project <YOUR_GCP_PROJECT_ID>
python3 -m venv .venv
source .venv/bin/activate
pip install google-adk

3. Create the ADK Project

Use the standard ADK project layout:

• a Python entrypoint for the agent
• a dependency file such as requirements.txt or pyproject.toml
• a local config file for environment variables and secrets

Keep the MCP connection details out of source control.

4. Connect the Demandbase MCP Server

Configure the ADK McpToolset to point at the Demandbase MCP server.

Use these connection details from Demandbase:

• MCP server URL: https://gateway.demandbase.com/mcp/servers/db-mcp
• Protocol: Streamable HTTP
• Auth method: Bearer token in the Authorization header

The MCP server exposes tools such as:

• search_internal_account_database
• search_internal_person_database
• company_global_directory
• contact_global_directory

Example agent wiring:

from google.adk.agents import LlmAgent
from google.adk.tools import McpToolset

db_mcp = McpToolset(
    server_url="https://gateway.demandbase.com/mcp/servers/db-mcp",
    headers={
        "Authorization": f"Bearer {MCP_BEARER_TOKEN}",
    },
)

agent = LlmAgent(
    model="gemini-2.0-flash",
    name="demandbase_agent",
    tools=[db_mcp],
)

5. Deploy to Vertex AI Agent Engine

Deploy the agent with the ADK CLI or your preferred deployment workflow.

1. Package the agent and dependencies.
2. Deploy to Vertex AI Agent Engine.
3. Capture the resulting reasoningEngine resource name.
4. Verify the deployment in the Google Cloud Console.

After deployment, confirm the agent can reach the Demandbase MCP server and list tools successfully.

6. Configure Gemini Enterprise

Register the deployed agent inside Gemini Enterprise.

1. Open Google Cloud Console.
2. Go to Vertex AI, then Agent Builder, then Gemini Enterprise.
3. Open the Gemini Enterprise app you want to extend, or create one.
4. Select Agents, then + Add agent.
5. Choose Custom agent via Agent Engine.
6. Paste the reasoningEngine resource name from Vertex AI Agent Engine.
7. Add authorization credentials required by the Workspace integration.

The Demandbase support team provides the authorization values:

• clientId
• clientSecret
• tokenUri
• authorizationUri

7. Verify the Integration

After saving the Gemini Enterprise configuration:

1. Open Gemini as a Workspace user.
2. Confirm the custom agent is available.
3. Run a request that should invoke a Demandbase MCP tool.
4. Check Vertex AI logs if the agent does not connect successfully.

Useful verification points:

• the agent returns tool results from Demandbase
• Gemini can authenticate the Workspace user flow
• the deployed Agent Engine resource is reachable from Gemini Enterprise

8. Troubleshooting

• If the agent cannot reach the MCP server, confirm the bearer token and server URL.
• If Gemini registration fails, confirm the OAuth credentials and the reasoningEngine resource name.
• If the agent deploys but tools do not appear, validate the McpToolset wiring in the ADK project.
• If requests fail after registration, review Cloud Logging for Vertex AI Agent Engine and the agent runtime logs.

9. Recommended Rollout Checklist

• GCP project ready
• Vertex AI enabled
• ADK environment created
• Demandbase MCP credentials configured
• Agent deployed to Vertex AI Agent Engine
• Gemini Enterprise authorization configured
• Agent visible to Workspace users
• End-to-end tool call verified