Dynamic (input) schemas in tools.

[Vanja-S]

Pre-submission Checklist

• I have checked that this question would not be more appropriate as an issue in a specific repository
• I have searched existing discussions and documentation for answers

Question Category

• Protocol Specification
• SDK Usage
• Server Implementation
• General Implementation
• Documentation
• Other

Your Question

Context and our situation

Unlike many other MCP providers, the tools we provide are dynamic, and change per-tenant basis, meaning different customers can have different inputSchemas for the same tools. To get those schemas, we call an internal microservice to retrieve them.

This retrieval can be quite costly, anywhere from 10-16s if there is a cache hit (the reasons why this cannot be optimized further on that side is beyond the scope of this question), so some MCP clients will timeout on tools/list.

The implementation currently solves this using partial tool returns, returning whatever tools have been processed, and sending an additional notification that the list has been updated, when the whole set becomes available (this is then always a cache hit).

Questions

1. In the scope of MCP is such an approach appropriate?
2. Is there a better way to utilize resources (as those input schemas) and reference them in tools input schemas?
3. Should we consider supporting tools with dynamic schemas which would need to be returned separately from the tools, say due to longer computation time?

[Vanja-S]

Moving to https://github.com/modelcontextprotocol
modelcontextprotocol