Google Cloud announced the public preview of a fully managed Model Context Protocol (MCP) server, initially covering BigQuery, Maps, Compute Engine, and Kubernetes Engine. Developers can simply paste the hosted endpoint URL to let agents like Gemini directly call Google services, without building connectors or maintaining credentials, achieving "zero-configuration" integration.
Initial Services: BigQuery + Maps Ready to Use
- BigQuery MCP: Agents can execute SQL in real time and return analysis results, used for data Q&A and automatic report generation
- Maps MCP: Replaces model internal knowledge, providing real-time location, route, and traffic information, supporting trip planning and store search
- Compute/Kubernetes MCP: Supports querying instance status and restarting Pods, used for operations inspection and self-healing from failures
Security and Governance: Dual Protection with IAM + Model Armor
- IAM Permissions: Fine-grained to table-level and cluster-level, agents can only access authorized resources
- Model Armor: Dedicated firewall, intercepting advanced threats such as prompt injection and data leakage
- Audit Logging: Fully records the agent's call trace, facilitating compliance audits
Enterprise Path: Apigee One-click "MCP Conversion"
- Existing APIs: Through the Apigee console, generate MCP endpoints with one click, making internal systems such as product catalog, orders, and CRM immediately become "agent tools"
- Quotas and Rate Limiting: Reuse existing API gateway policies, no additional development needed
- Cost: Free during the public preview, the final version will be included in the existing cloud bill, with no additional connector fees
Release Schedule: Weekly Additions, Full Product Line Coverage by 2025
- Next Week: Logging, Monitoring, and Cloud Storage MCP will go live
- Q1 2025: Database, Security, and IAM Suite will be fully MCP-enabled
- Long Term: Average of 1-2 new service MCP releases per week, covering 80% of Google Cloud services by year-end
Industry Significance: MCP Evolves from "Protocol" to "Infrastructure"
Google upgraded the MCP from an open-source protocol to a managed infrastructure, meaning:
- Developers: No need to maintain connectors, tokens, or version compatibility, focusing on business logic
- Enterprises: Existing API governance, quotas, and audit systems can be directly reused for AI agents, reducing compliance risks
- Ecosystem: Any MCP-compatible client (Gemini, Claude, ChatGPT) can call it, achieving "one-time access, multi-end compatibility"
Next Steps: General Hosting + Edge Nodes
