Managed MCP servers for Google Cloud databases | Google Cloud Blog
Google Cloud launches managed MCP servers for their database portfolio, letting AI agents directly interact with PostgreSQL, Spanner, Cloud SQL, Firestore, and Bigtable through the Model Context Protocol standard. No infrastructure to deploy — just configure endpoints and agents get secure, governed access to operational data.
Script: Sonnet 4.5 Voice: OpenAI TTS
Transcript
Izzo Google just solved the agent-database connection problem in the most Google way possible.
Izzo You're listening to Exploring Next, episode two-oh-four. I'm Izzo, here with Boone, and today we're talking about something that's gonna make a lot of AI developers very happy.
Boone Managed MCP servers for Google Cloud databases. And Izzo, this is actually brilliant architecture.
Izzo Right? Because here's the thing — everyone's building agents now, but connecting them to your actual operational data has been this huge pain. You're writing custom APIs, managing auth, worrying about security.
Boone Exactly. And Google just said 'what if you didn't have to do any of that?' They're running the MCP servers for you.
Izzo Walk me through what MCP actually is, Boone. I know it's the Model Context Protocol, but what does that mean in practice?
Boone Think of it as a universal translator between AI agents and tools. Instead of every agent needing custom code to talk to PostgreSQL or Spanner, MCP creates this standard interface. The agent speaks MCP, the database speaks MCP, boom — they understand each other.
Izzo And Google's managing the servers that do this translation?
Boone Yep. You just configure an endpoint in your agent, and suddenly it can query your AlloyDB instance, run vector similarity searches, even create schemas. No infrastructure to deploy.
Izzo Okay, that's the product story I was hoping for. Because the alternative is every developer building their own database connectors for every agent project.
Boone Which is exactly what's been happening. And it's messy and insecure.
Izzo So what databases are we talking about here? The whole Google Cloud portfolio?
Boone Pretty much. AlloyDB for PostgreSQL gets schema creation and vector search. Spanner gets the really interesting stuff — since it's multi-model now, agents can query relational data and graph data in the same operation.
Izzo Wait, hold on. Spanner can do graph queries now?
Boone Spanner Graph, yeah. So imagine an agent that's investigating fraud — it can use SQL to pull transaction data and GQL to map the relationship networks, all in one query through the MCP server.
Izzo That's... actually pretty powerful. What about the NoSQL side?
Boone Bigtable for high-throughput time series data, Firestore for mobile and web app data. The Firestore one's interesting — agents can sync with live document collections, check user session states, verify order status.
Izzo So you could build a customer service agent that actually knows if someone's order shipped?
Boone Exactly. And it's all through natural language prompts.
Izzo I'm giving this a solid A-minus just for the product vision. But what about security? Because connecting agents to production databases makes me nervous.
Boone They thought about that. Everything goes through IAM instead of shared keys. So the agent can only access tables and views you explicitly authorize.
Izzo Smart. No more API keys floating around in config files.
Boone And full audit logging in Cloud Audit Logs. Every query, every action gets recorded.
Izzo Governance teams are gonna love that. But here's what I really want to know — does this actually work with non-Google agents?
Boone That's the best part. It's the open MCP standard, so you can connect Claude, or any MCP-compliant client. Just add a custom connector pointing to your Google Cloud endpoint.
Izzo So Google's not trying to lock you into Gemini. Interesting strategy.
Boone They're betting that making their databases the easiest to connect to will win in the long run. And honestly? That's probably right.
Izzo There's also this Developer Knowledge MCP server that connects IDEs to Google's documentation. That feels like a different category of tool. Yeah, but it makes sense in context. Your agent's migrating data to Cloud SQL, hits a configuration question, and boom — it can query Google's docs directly for the answer. Instead of the developer having to context-switch to Stack Overflow. I like it. I'm definitely adding this to the weekend project list. Want to see how fast I can get