Goal

This issue aims to explore and assess the performance and scalability of the MCP server within the ToolHive ecosystem when subjected to increased load and concurrency. The objective is to identify how the system behaves under stress, what bottlenecks emerge, and where we may need to improve architecture or runtime isolation.

Key Questions

• What is the limit for stdio, sse and streamable-http servers in terms of concurrent requests per second?
• How do the servers perform under consistent load?
• Are there bottlenecks? If so, where?

Deliverables

• Summary of performance metrics and failure patterns.
• Table of test scenarios and outcomes.
• Highlighted bottlenecks or scaling concerns.
• Recommendations for architectural or runtime improvements that spawns future work (if there are any).