Expired60% confidenceFactExact time
MCP's default implementation injects the schemas of all registered tools into the context at session initialization; when the number of tools reaches dozens or even hundreds, tool descriptions alone may consume thousands of tokens.
2
Sources
60%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
Valid until: 9/30/2026(expired)
Sources
What Are Agent Skills? A Deep Dive into the New Paradigm for AI Agent Decision-Making
bilibiliAI大模型系统课程6/25/2026
Related Claims
Unverified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token84% similarVerifiedWhen connecting to multiple MCP servers simultaneously, tool descriptions alone can consume thousands or even tens of thousands of tokens.77% similarUnverifiedThe scalability problem with MCP is that the name, description, and schema of each tool must be placed in the system prompt, and 20 servers with 15 tools each would occupy most of the context window.76% similarUnverifiedMCP's design philosophy is full registration — tool descriptions from all servers are injected into the context all at once during session initialization.76% similarUnverified工具调用已从单次简单查询演进为支持并行工具调用、工具链式编排以及MCP等标准协议的复杂多步骤编排系统75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/41085API
curl https://kongchang.com/api/v1/knowledge/claims/41085MCP
get_claim(id=41085)