Microsoft is expanding its AI platform by adding support for the Model Context Protocol (MCP), a vendor-neutral system that lets AI agents exchange memory and tools over a shared interface.
Originally introduced by Anthropic in late 2024, the MCP enables structured communication between agents and services using a simple HTTP schema. Microsoft’s implementation brings MCP into Azure AI Foundry, complete with integration in Azure AI Agent Service and a new C# SDK.
The move positions Azure as a more open and interoperable platform for multi-agent workflows. Developers can now build agents that consume MCP servers for memory, tools, and data, or expose those capabilities to others using standardized schemas.
The MCP allows AI agents to use tools and memory over a shared schema and supports cross-vendor interoperability, allowing agents built with different models or frameworks to interact with the same services.
Replacing Fragmented Integrations with Open Standards
The MCP operates using a client-server model. AI agents act as clients that connect to MCP servers, which expose tools and memory via typed interfaces. Each tool endpoint defines structured input and output schemas, while memory APIs provide context persistence.
Communication occurs over standard HTTP, making it easy to deploy across environments ranging from local machines to cloud services.
Microsoft’s deployment follows this model closely. Its Azure integration leverages FastAPI-based server templates and Docker configurations maintained in the official GitHub repository.
These examples let developers create task-routing agents, trigger cloud APIs, or manage persistent memory states using a shared schema. MCP clients can call these tools, pass parameters, receive structured outputs, and access memory to maintain coherent task execution.
Support extends beyond Azure AI Agent Service. Microsoft has also integrated the MCP into its Semantic Kernel framework, letting developers build workflows that connect models to real-time data from Bing Search or internal data from Azure AI Search.
The protocol is open source and hosted under the modelcontextprotocol GitHub. SDKs are available in Python, TypeScript, Java, Kotlin, and now C#. The shared MCP specification ensures compatibility across implementations.
Anthropic’s Push for Scalable Agent Communication
Anthropic introduced the MCP in November 2024 to address a growing complexity in AI systems: agents couldn’t easily interact with external tools or share memory across different platforms.
As the company explained at launch, “Every new data source requires its own custom implementation, making truly connected systems difficult to scale.”
The protocol quickly became a key part of Anthropic’s infrastructure. Claude Desktop was shown using the MCP to perform developer tasks such as creating GitHub repositories and opening pull requests.
“Once MCP was set up in Claude desktop, building this integration took less than an hour,” wrote Anthropic engineer Alex Albert, describing a GitHub workflow that used MCP tools to interact with the file system and local shell.
Here's a quick demo using the Claude desktop app, where we've configured MCP:
— Alex Albert (@alexalbert__) November 25, 2024
Watch Claude connect directly to GitHub, create a new repo, and make a PR through a simple MCP integration.
Once MCP was set up in Claude desktop, building this integration took less than an hour. pic.twitter.com/xseX89Z2PD
Since then, Anthropic has expanded MCP support to include JavaScript execution, persistent memory, and automation capabilities. Replit, Sourcegraph, Apollo, and Block are among the early adopters that now use the MCP to build agent tools and connect cloud systems to locally hosted models.
Anthropic’s infrastructure strategy is backed by major funding. Amazon has invested $4 billion in Anthropic to train Claude on its custom Trainium and Inferentia chips and to serve it via Amazon Bedrock. Google also invested with a $2 billion equity stake. On November 19, 2024, the UK Competition and Markets Authority ruled that the deal would not give Google material influence over Anthropic.
How The MCP Fits Microsoft’s Open AI Strategy
Microsoft’s support for the MCP is part of a broader realignment of its AI infrastructure around modularity and interoperability. In January 2025, the company launched a new division, CoreAI – Platform and Tools, led by former Meta executive Jay Parikh.
The group consolidates Azure, GitHub, and developer platform teams to accelerate work on systems like Azure AI Foundry and cross-model agent tooling.
That strategy includes expanding the model offerings within Azure. Microsoft later added the Chinese open-weight DeepSeek R1 reasoning model to Azure AI Foundry and GitHub. The model supports code generation and document reasoning and provides a cost-effective alternative to models like GPT-4.
Developers can now build workflows using R1 in combination with MCP tools, aligning with Microsoft’s move to support multiple model providers in a shared agent environment.
With the MCP in place, Azure Foundry becomes a platform for building agentic AI systems that span clouds, models, and data sources. Instead of relying solely on model-specific APIs like OpenAI function calling, developers can now use an open schema to orchestrate tools and memory across components.
Technical Trade-Offs and Adoption Questions
While the MCP’s architecture offers flexibility, it also introduces some technical trade-offs. Its reliance on HTTP makes integration easy but may introduce latency in real-time or high-frequency applications. The protocol’s generality also places more responsibility on developers to build caching, error handling, and security layers appropriate for production workloads.
Microsoft’s official support for the C# SDK adds enterprise credibility to the ecosystem, especially for .NET environments. However, many of the other SDKs—like those in Python and TypeScript—remain community-maintained, which could slow adoption in regulated industries where long-term support guarantees are essential.
Even so, the MCP is gaining traction. The official GitHub repository shows regular contributions and collaboration between multiple stakeholders, including Microsoft and Anthropic. The presence of mature examples and deployment templates further lowers the barrier to experimentation and deployment.
For developers building AI agents that need to reason, retrieve, and act across multiple sources and services, the MCP offers a shared vocabulary and an emerging standard. With Azure AI now supporting it across multiple services, the protocol moves one step closer to becoming a foundational layer in the architecture of model-agnostic AI systems.
