Google Cloud

This guide covers deploying the Falcon MCP Server with the prebuilt Google ADK-based agent to Cloud Run or Vertex AI Agent Engine.

Prerequisites

• Python 3.11+, gcloud CLI, and git installed
• Google Cloud project with billing enabled
• CrowdStrike API credentials
• GOOGLE_API_KEY from Google AI Studio

Running Locally (5 minutes)

Clone the repository:

git clone https://github.com/CrowdStrike/falcon-mcp.git
cd falcon-mcp/examples/adk

Create Python environment and install dependencies:

pip

python3 -m venv .venv
. .venv/bin/activate
pip install -r falcon_agent/requirements.txt

uv

uv venv
. .venv/bin/activate
uv pip install -r falcon_agent/requirements.txt

Initialize config:

chmod +x adk_agent_operations.sh
./adk_agent_operations.sh

The script creates a .env file in falcon_agent/. Update at minimum the General Agent Configuration section with your CrowdStrike credentials and Google API key.

# Run locally
./adk_agent_operations.sh local_run

Access the agent at http://localhost:8000.

Caution

Do not use curly braces ({variable}) in the FALCON_AGENT_PROMPT value. Google ADK interprets {name} patterns as context variables, which causes Context variable not found errors at runtime.

Deploying to Cloud Run

Make sure the required APIs are enabled on your GCP project.

cd examples/adk/
./adk_agent_operations.sh cloudrun_deploy

When prompted Allow unauthenticated invocations?, answer N to keep IAM authentication enabled.

Once deployed, grant access to your team:

1. Cloud Run > Services > falcon-agent-service > Permissions
2. Add Principal > assign Cloud Run Invoker role

Your team members can then access the service via:

gcloud run services proxy falcon-agent-service --project PROJECT-ID --region YOUR-REGION

The service is then available at http://localhost:8080.

Deploying to Vertex AI Agent Engine

Create a GCS bucket for staging artifacts, then:

cd examples/adk/
./adk_agent_operations.sh agent_engine_deploy

Note the Agent Engine Number from the output (reasoningEngines/XXXXXX).

Registering with Agentspace

Update the # Agentspace Specific environment variables (PROJECT_NUMBER, AGENT_LOCATION, REASONING_ENGINE_NUMBER, AGENT_SPACE_APP_NAME), then:

./adk_agent_operations.sh agentspace_register

FQL Guide Resources

The agent is configured with use_mcp_resources=True, enabling ADK’s MCP resource support. The Falcon MCP Server exposes FQL guide resources (e.g., falcon://detections/search/fql-guide) that the agent fetches on demand via the auto-discovered load_mcp_resource tool, providing accurate Falcon query construction without embedding all FQL documentation in the system prompt.

Performance Optimization

Control MAX_PREV_USER_INTERACTIONS in your .env to limit conversation history sent to the LLM (recommended: 5). This reduces costs while maintaining useful context.