ICML Paper Search MCP
Deployment and Operations

Capacity

Default limits for the public deployment.

LimitDefault
Active sessions4,096
Search queue512
DashScope calls64 concurrent
Milvus calls32 concurrent
Per IP600 requests/minute, burst 100
Global application rate6,000 requests/minute
Provider rate24,000 requests/minute
Provider input1,000,000 tokens/minute

These values are intended for roughly 25–40 simultaneous users with short search queries. Nginx, the application, and the provider quota are separate controls. Measure queue delay, provider latency, Milvus latency, SQLite lock time, and memory before increasing them.