The outcome: in about 5 minutes, your AI agent keeps every tool it has today but stops re-paying for them on every message. Same stack, 96 to 99 percent fewer wasted tokens. Measured, not estimated.
The 60-second why
Every MCP server you connect injects its full tool catalog into your agent's context on every single message, whether the tools get used or not.
- ~121 tokens per tool, per message (native MCP schema injection)
- 30 tools = ~3,600 tokens per message before the agent does anything