An Agent Can't Offer an MCP Tool It Can't See
MCP discovery tells an agent how to authenticate, not what a tool does. When a tool sits behind per-user login, you need static metadata. A smarter model will not help.
~ % grep -r "agents" ./blog
3 posts found
MCP discovery tells an agent how to authenticate, not what a tool does. When a tool sits behind per-user login, you need static metadata. A smarter model will not help.
Hard limits for AI agents belong in code. The model reads prompt rules and breaks them. A deny-by-default gate with no way to say yes fails too.
AI made finding code issues nearly free. Owning the fix still needs a human. Finding throughput is a vanity metric; the hand-back is the real work.