Get more out of AI agents connected to your workspace via the MCP server

Since its release in November 2025, we have substantially updated our MCP server. AI agents and custom applications connected via the MCP server — whether through Claude, Microsoft Copilot Studio, Cursor, or other MCP-compatible clients — now have access to a richer and more reliable toolset.

New capabilities

Querying applications now better supports filtering by lifecycle phase and fact sheet subtype, making it straightforward to ask questions like: “Which active microservices are approaching end of life?” Additionally, all tools now return direct links, so agents can point users straight to the relevant item in a workspace.

Stability, security, and bug fixes

Core operations are now substantially more stable, errors are properly surfaced to agents, and we have refined tool descriptions for better compatibility with Microsoft Copilot Studio. We also increased rate limits to support agent workloads with high tool-call volumes. If you use technical users with access control entities within virtual workspaces, you can now connect to the MCP server without restrictions. Lastly, field-level visibility rules are now respected: hidden fields are fully omitted from responses rather than returned with empty values, preventing agents from inadvertently exposing restricted field names.

What’s next?

As announced in our product roadmap, we are working to open the MCP server to all users by enabling user-based access via OAuth. After that, we will allow agents to perform inventory data changes (create, update, and delete) via the MCP server.

Want to contribute?

With each update, we run proprietary automated evaluations (EVALs) to test the impact of changes on the accuracy and reliability of agent responses. If you have created skills, agents, or scripts running against our MCP server and want us to include respective EVALs in our pipeline, please reach out to us via the above roadmap item.