Exposing existing business systems over MCP

[yinzara]

Pre-submission Checklist

• I have verified this would not be more appropriate as a feature request in a specific repository
• I have searched existing discussions to avoid duplicates

Your Idea

I've been attempting to build MCP servers to expose existing complex business systems to our LLM to help our operators and users interface with the system. Unfortunately I keep using up most of the context window of Claude just describing my toolset because I have to repeat the same information over and over again both in inputSchemas and outputSchemas. Business objects are often predictable and carry very common properties and value objects that when you have some kind of ubiquitous schema to refer between becomes very small. Unfortunately because MCP has no way of referring to the same schema between tools or even between the input and output schema of a specific tool, there is no way to optimize the context window used by these MCPs.

For example, I have a value object called "LocalizedValueDTO" that contains a "value" string property and an "i18n" dictionary property that allows me to represent a value and its translations on a property of a business object. I use this value object across more than 100 business objects sometimes on multiple properties of the same object. Because I have no ability to refer to some ubiquitous schema between tools, every time I use a LocalizedValueDTO, I have to fully describe it using the JSON Schema specification. This blows up my context window by so much repeating the same thing over and over again because I can't be more succinct.

Help me!

I opened an issue to propose a fix for this:
modelcontextprotocol/modelcontextprotocol#610

but no one from Anthropic has responded.

Scope

• Protocol Specification
• SDK Features
• Documentation
• Developer Experience
• Other