Microsoft has shipped the MCP Toolkit for Azure Cosmos DB as generally available (v1.1.2), turning your operational database into an AI-agent-accessible endpoint through the Model Context Protocol. [source]
The toolkit acts as an MCP server that sits between your AI agents and Cosmos DB, handling authentication, query translation, and vector search without custom SDK wrappers. It supports document CRUD, schema discovery, vector and hybrid search, and Azure Entra ID authentication out of the box. [source]
What’s new in GA
The GA release adds multi-provider embedding support (Azure OpenAI, Azure AI Services, Foundry project endpoints) via a new abstraction layer that auto-detects your endpoint type. Error handling has been overhauled for production workloads, and the toolkit now integrates directly with Microsoft Foundry for agent orchestration. [source]
Why this matters
Until now, connecting an AI agent to a production database meant custom code, security reviews, and brittle integration layers. The MCP toolkit standardizes that interface. If you are running Cosmos DB and building AI agents, this removes most of the plumbing work. You point the toolkit at your database, configure an embedding provider, and your agent can query, search, and reason over your data using natural language. For teams already in the Azure ecosystem, the Foundry integration means you can wire this up without leaving the platform.
Getting started
- Deploy the toolkit as a Docker container in Azure Container Apps, AKS, or any container runtime. [source]
- Configure Azure Entra ID authentication and RBAC at the container level.
- Point your AI agent (Foundry, LangChain, or any MCP client) at the toolkit endpoint.
- Available tools include listDatabases, queryContainer, searchVectors, and schema discovery. [source]
This is one of those releases that quietly makes a lot of things simpler. If you have been avoiding the agent-to-database integration problem, the excuse just got thinner.