Guides
Agents
Putting agents on your data: MCP, custom tools and guardrails, and the approvals queue.
MCP is how AI agents operate your data. These pages cover the tool surface itself, the policy layer in front of it, and the human approval step for calls a policy can't auto-allow.
- MCP — expose objects as agent tools, per identity
- Custom tools, guardrails & audit replay — the open contract
- Approvals — the human-in-the-loop queue
- Execution & evidence — the shared write/tool pipeline and its evidence record