-
Notifications
You must be signed in to change notification settings - Fork 103
Description
Problem
On startup Newelle loads MCP servers, tools etc into the next window on load which very quickly slows down the response times and massively increases token usage
Despite being able to selectivley turn off specific tools for each MCP server (which is great) if you have serveral MCP servers / agents it very quickly becomes unmanageable
Proposed solution
Implement lazy loading for MCP servers and tools, loading them only when needed based on conversation context, potentially based on this implementation, which was integrated into Claude Code only 5 days ago
Benefits
95% Token Reduction
Longer Conversations
Better Performance: Faster startup, lower memory usage
Scalability: Can add more tools without context penalty
Related improvements
Can we add 'per profile' Tools settings please? e.g
Profile 1: MCP servers A, B
Profile 2: MCP servers C, D
Thank you once again for Newelle, great work!