Articles
How to design the parts specific to AI products, such as MCP consent screens, tool approval and agent runs. Each article comes with a live component that implements it.
- Ask before the agent actsHow an agent should ask for permission before it sends, deletes or spends: which calls to ask about, what the question should say, and what "always allow" can safely mean.
- What an MCP consent screen needsWhen an agent connects to your MCP server, the consent screen is the only thing between a person and a client they've never heard of. Eight things it has to get right, with a working one to try.