Nice work, a web-based MCP host feels like the missing middle ground between local desktop clients and hand-rolled scripts.

Have you noticed any practical limits once users connect 4–5 MCP servers at the same time (latency, tool selection confusion, context bloat)?

Would be interesting to hear what breaks first in real usage.