Expired65% confidenceFactExact time
When each MCP is loaded, its tool descriptions are stuffed into the context; attaching a large number of MCPs to an Agent without restricting Tools will cause token explosion, increased latency, and rising costs.
3
Sources
65%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
Valid until: 9/30/2026(expired)
Sources
The Right Way to Use Claude Code Sub-Agents: 3-5 Beat 100 Every Time
bilibili土木ai提效大古6/3/2026
Related Claims
UnverifiedIn high-frequency invocation scenarios, MCP costs scale exponentially due to persistent tool description injection79% similarUnverifiedAt the tool layer, using large models with Function Calling or the MCP protocol to handle complex requests can only serve as a fallback and cannot carry the main traffic; otherwise, costs will severely overrun as user volume grows.77% similarUnverifiedMCP的工作机制需要将大量上下文信息注入模型输入，导致Token消耗过大75% similarUnverified建议在自定义MCP工具实现中主动向代理返回操作后的相关日志摘要，将两次工具调用合并为一次，以压缩迭代轮次并降低token消耗与延迟72% similarUnverified每个MCP工具的定义描述、参数schema、使用示例等加在一起可能消耗数百乃至数千token72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/52252API
curl https://kongchang.com/api/v1/knowledge/claims/52252MCP
get_claim(id=52252)