AI agents possess incredible reasoning capabilities and can perform increasingly complex actions. But the reliability of agentic outcomes depends entirely on the quality of the context they can access** **— context that is frequently locked away in operational databases.
To bridge this gap, we are excited to announce the Remote Model Context Protocol (MCP) Server for AlloyDB is now generally available.
The Model Context Protocol (MCP) is an open-source standard that gives LLMs a secure, consistent way to connect to external data sources. As part of Google Cloud’s recent rollout of 50+ Google-managed MCP servers, this new integration makes it easier than ever for both interactive and autonomous agents to securely harness the full power of your enterprise data. For example, you can now ask an AI agent for an up-to-the-millisecond view of your delivery fleet by connecting it to your real-time logistics data in AlloyDB, avoiding inaccuracies due to stale data and reducing the need for manual reporting.
By connecting MCP to AlloyDB, your agents get access to the premier database built for enterprise-grade AI. AlloyDB delivers the scale, speed, and intelligence required for the most demanding agentic workloads:
Supercharged vector performance: Scale to over 10 billion vectors at up to 6x the speed of standard PostgreSQL for vector queries (and up to 10x faster for filtered queries) with the ScaNN index.
Advanced search and reranking: Power multimodal applications with hybrid search via RUM (in Preview) and intelligent reranking through Reciprocal Rank Fusion (RRF) or Gemini Enterprise Platform models.
Real-time intelligence: Efficiently generate millions of embeddings using built-in AI Functions to facilitate low-latency, real-time agentic experiences.
Unified data access: Give agents a single PostgreSQL interface to seamlessly join operational data in AlloyDB with analytical data in BigQuery or archived data in Iceberg tables via Lakehouse Federation.
Enterprise-grade scale: Rest easy with a 99.99% SLA, autopilot database optimizations, and auto-scaling read pools with up to 20 nodes.
Local MCP servers are great for local development, but communicating over standard input/output (stdio) streams becomes difficult when you scale to production workloads. It is both architecturally complex and administratively burdensome to provision and manage all of the infrastructure and security guardrails you need to run agents for high-value use cases that interact with sensitive operational data.
The Remote MCP Server for AlloyDB runs on fully-managed Google Cloud infrastructure and exposes an HTTP endpoint that connects your AI applications to your data. This solves key challenges for teams building agents on PostgreSQL:
Centralized discovery: Find, secure, and manage your database's MCP server using Agent Registry.
Fully-managed HTTP endpoints: No need to deploy or maintain the infrastructure required for connectivity. Configure your agent to use the endpoint to get started.
Fine-grained authorization: Instead of using shared database passwords or API keys, you use Identity and Access Management (IAM) to restrict agents to specific tables, schemas, or views. With the read-only execute SQL tool, you can prevent your agent from making accidental changes and deletions from your database.
Operational instance management: The AlloyDB toolset gives agents the ability to do more than run queries. Agents can update instances, export and import data, create backups, and restore clusters.
Model Armor protection: Model Armor provides optional prompt and response security to screen and filter data, defending against prompt injections or accidental data exfiltration.
Audit logging: Every query, action, and tool call goes to Cloud Audit Logs, giving security teams a full audit trail.
Getting started with the AlloyDB Remote MCP server is a straightforward process. To see it in action in your own environment, you can follow our new Codelab, which guides you through these essential steps:
API & environment prep: Enable the AlloyDB, Compute Engine, and Gemini Enterprise APIs in your Google Cloud project.
Provision your database: Deploy your AlloyDB cluster, create your database, and import your sample data.
Enable data access API: Permit the Data Access API on your AlloyDB instance.
Connect the agent: Configure your MCP client by providing the remote endpoint (https://alloydb.googleapis.com/mcp
). Pass your Google Cloud IAM credentials using an OAuth 2.0 bearer token in the HTTP Authorization header.
Once the connection is established, your agent can provide reliable, grounded answers to complex business questions using your real-time operational data. By performing introspection queries, the agent automatically understands your database schema – including tables and columns – enabling it to construct sophisticated joins and queries to fulfill user requests accurately.
Once your agent has access to the AlloyDB toolset, it can execute queries, analyze operational trends, and dynamically rank text data using AlloyDB AI functions like AI.RANK()
.
Security remains paramount: the Remote MCP Server for AlloyDB integrates seamlessly with Model Armor. This provides protection against sensitive data leaks, even if the agent’s service account possesses broad access permissions within the database.
Watch the full demo below!
By enabling agents to interact securely with transactional data, we are embracing an architecture where AI agents can reliably access and act upon your enterprise’s single source of truth.
Ready to build? Discover AlloyDB with a 30-day free trial, and dive into the Remote MCP for AlloyDB Codelab to start powering your enterprise agentic applications today.