How to Use Google’s Managed MCP Servers on Google Cloud
Google rolled out managed Model Context Protocol (MCP) servers this week for AI agents to connect directly to its cloud services. MCP, an open standard from Anthropic now under the Linux Foundation’s Agentic AI Foundation, acts like a standard connector for agents to reach tools and data. No more custom patches—developers just point their agent to a managed endpoint URL and go. As Steren Giannini from Google Cloud told Yahoo Tech, setup drops from weeks to minutes.
Available Services at Launch
Right now in public preview, you get MCP servers for these:
- Maps—for live location data in agents like trip planners.
- BigQuery—for direct queries from analytics agents.
- Compute Engine—for infrastructure management.
- Kubernetes Engine—for cluster operations.
More roll out soon, hitting storage, databases, logging, monitoring, and security. Google plans weekly additions, per Startup News FYI, SDxCentral and FindArticles.
Who Can Use Them
Enterprise customers already paying for these Google Cloud services get them at no extra cost. It’s public preview, so not fully under standard terms yet—general availability hits early next year. Works across clients like Gemini CLI, AI Studio, Claude, and ChatGPT. Google tested it themselves, as noted in the Yahoo Tech piece.
Quick Setup for Basic Use
- Get your project ready: Log into Google Cloud Console with a project that has billing and access to the target service (e.g., enable BigQuery API).
- Grab the MCP endpoint: In the service’s console or docs, find the managed MCP server URL. Paste it into your agent’s config—your LLM client discovers and calls tools from there.
- Test a connection: Fire up Gemini CLI or AI Studio, add the URL, and query something simple like BigQuery data or Maps info. Agents get fresh, grounded responses instead of model guesses.
That’s the core flow, straight from Google’s demos in FindArticles.
Set Permissions with IAM
Control access via Google Cloud IAM. Assign least-privilege roles to your agent—what it can query or manage on that MCP server. Layer on Model Armor for defenses against prompt injection or data leaks, plus audit logs for tracking. Giannini called it a dedicated agent firewall in Yahoo Tech.
Integrate Custom APIs with Apigee
Got your own APIs? Use Apigee to turn them into MCP servers. It handles keys, quotas, and monitoring, so agents treat them like native Google tools. Same guardrails apply—no new security setup needed. Details in the FindArticles coverage.
Watch the Standards Push
MCP’s rise got a boost with Anthropic handing it to the Linux Foundation’s AAIF, alongside firms like OpenAI and Block. Google added support in its tools post-I/O 2025, per Ars Technica. Expect broader compatibility. A short note on the launch appeared via Bitcoin Ethereum News.
Agents finally plug in reliably. Start small with BigQuery or Maps, then scale.
