Agentic, Multimodal, and Real-Time Applications

MCP supports agentic systems in which models decide which tools to use. A model can inspect available capabilities, select a tool, receive the result, and continue the task. This is useful for workflows that require more than one external action or data source.

A model can work with one server or many servers. Each server can provide a different capability, and the agent can select among them. This allows a single application to combine data stores, customer systems, search, version control, calculations, and other functions.

Multimodal implementations extend the same architecture beyond text. An MCP server can support images, audio, and generated media summaries. This enables richer applications that combine several forms of information.

Real-time access is another important use case. Servers can connect models to current documentation, market information, search, enterprise systems, and external APIs. This gives the model context that was not contained in its training data and can reduce unsupported answers by grounding the response in retrieved information.

Streaming supports real-time interaction. Server-sent events, HTTP streaming, and WebSocket streaming are used in the examples to return information progressively.

Enterprise integrations include Azure OpenAI and Microsoft AI Foundry. These integrations support tool orchestration, real-time web search, external API connections, and identity and access management.

Advanced samples also cover routing and sampling strategies. A multi-provider MCP server can route requests to different model providers according to metadata, allowing OpenAI, Anthropic, and local models to operate behind one server design.