Use the MCP server with per-user access and custom meta model support

We have released three improvements to our MCP server, which make it easier to deploy AI agents across your organization at scale. The updates address access management, meta model accuracy, and performance.

User-based access

AI agents using the MCP server can now authenticate on behalf of individual users via OAuth, rather than relying on shared tokens or manual permission management. You can roll out the MCP server to your entire organization without additional administrative overhead. Each user’s agent sessions automatically reflect their existing LeanIX permissions, ensuring architecture data is accessed securely and in line with individual access rights.

Better custom meta model support

The MCP server now understands your workspace’s configured meta model better. Custom fact sheet names, renamed fields, and tailored attribute labels defined in your meta model configuration are reflected in agent interactions. Rather than only relying on technical identifiers, AI agents interact with your architecture data using the terminology your organization has established, thereby improving the accuracy and relevance of AI-generated outputs.

Default and optional toolsets

The MCP server has been enhanced with a set of default toolsets. This allows us to continue adding more and more optional toolsets that are not automatically included in the scope of the connected LLM. This improves token efficiency and scalability, giving organizations better control over what is loaded for each agent session without unnecessary overhead.