MCP Architecture: Handling 160+ endpoints without context bloat

[renatomarinho]

Pre-submission Checklist

• I have verified that this discussion would not be more appropriate as an issue in a specific repository
• I have searched existing discussions to avoid duplicates

Discussion Topic

Hey everyone,

We just open-sourced our MCP server for GitScrum and hit a major bottleneck: exposing 160+ operations without wrecking Claude’s context window.

Mapping every endpoint to a single tool made the model hallucinate. We solved it by using a Consolidated Tool Pattern—grouping operations by domain (e.g., tasks, sprints) and using an action parameter. This kept the system prompt lean while maintaining full API coverage.

Question for the devs: When dealing with massive APIs, are you grouping tools like this or splitting them into multiple granular MCP servers?