Question 1

What is the Agent Approval Protocol?

Accepted Answer

AAP is withHuman’s open specification for how agents request, record, and obtain human approval before sensitive actions—see withhuman.ai/docs/reference/agent-approval-protocol.

Question 2

How is withHuman different from Agent Tool-Call Guard?

Accepted Answer

Tool-call guards classify whether a tool invocation is allowed. withHuman emphasizes human attestation and MCP gateway custody for exceptions and credential isolation—teams often deploy both layers.

Question 3

Does withHuman work with MCP only?

Accepted Answer

Documentation centers on MCP gateway patterns, but approval concepts apply to any tool-capable agent host—confirm supported transports on withhuman.ai.

Question 4

Can approvals happen on mobile?

Accepted Answer

Launch messaging includes mobile-friendly escalation paths alongside Slack—verify current clients in product docs.

withHuman: Agent Approval Protocol and MCP Gateway

Alternatives

GitHub Copilot

Cursor

Replit AI

v0 by Vercel

Bolt.new

Codeium