シングル呼び出しからエージェントへ：Microsoft FoundryでClaudeの5つの新機能が利用可能にFrom single call to agents: five new Claude capabilities now available in Microsoft Foundry

Microsoft has expanded the set of capabilities available to Claude models running inside Microsoft Foundry, adding five features that collectively move the service beyond simple request-and-response usage toward building autonomous agents. For teams already running Anthropic's Claude on Azure, the update matters because it reduces the amount of surrounding infrastructure they need to assemble themselves to take a model from a prototype to a production system.

According to Microsoft, structured outputs, web search, web fetch, an MCP connector, and tool search are now available for Claude models hosted on Azure in Microsoft Foundry. The company frames the change as turning a model endpoint into a production agent platform, which is the reasoning behind the "from single call to agents" description. Each of the five features addresses a different part of that transition, and together they target common gaps between a raw model call and a dependable application.

Structured outputs constrain a model's response to a predefined schema, typically expressed as JSON. This is aimed at reliability: when a model's output must feed directly into downstream software, databases, or APIs, free-form text is fragile and error-prone. Enforcing a schema makes the response easier to parse and validate, which is often a prerequisite for production workflows that cannot tolerate malformed data.

Web search and web fetch both address grounding, the practice of connecting a model to current information rather than relying solely on its training data. Web search allows the model to issue queries and incorporate results, while web fetch retrieves the contents of specific URLs. These tools help mitigate outdated answers and can support use cases where responses need to reflect recent events or reference material the model has not seen.

The MCP connector brings support for the Model Context Protocol, an open standard originally introduced by Anthropic for connecting AI models to external tools and data sources through a consistent interface. MCP has gained traction across the industry as a way to avoid building bespoke integrations for every data source, and its inclusion here suggests Foundry is aligning with that emerging convention rather than requiring a proprietary approach. In practice, an MCP connector lets a Claude-based agent reach servers that expose tools, files, or services in a standardized way.

Tool search addresses a scaling problem that appears once an agent has access to many tools. Loading every tool definition into a model's context consumes tokens and can degrade the model's ability to select the right one. Tool search appears designed to let the model discover and retrieve only the relevant tool definitions when needed, which is likely to help agents that operate over large tool catalogs remain efficient and accurate.

For context, Microsoft Foundry is the company's platform for building, evaluating, and deploying AI applications, and it hosts models from multiple providers alongside Microsoft's own tooling. Offering Anthropic's Claude on Azure fits a broader pattern in which cloud providers position themselves as neutral hosts for several model families, giving customers a choice of underlying models while keeping billing, security, and compliance within a single environment. The additions announced here are consistent with a wider industry shift toward server-side tools and agentic features being packaged with the model endpoint itself, rather than left entirely to application developers.

The move also reflects how quickly the notion of an "agent" has become central to enterprise AI. Structured outputs, retrieval from the web, standardized tool connections, and tool selection are recurring components in most agent frameworks, and providers increasingly compete on how much of that plumbing they supply out of the box. By bundling these five capabilities, Microsoft is reducing the integration burden for teams that want Claude to take actions, call external systems, and return machine-readable results rather than simply generate text.

As with any platform update, the practical value will depend on details such as regional availability, supported Claude model versions, pricing, and how the features behave under real workloads. Organizations evaluating the update would be well served to confirm those specifics against Microsoft's documentation before committing production systems to the new capabilities.