# subagentpermissions.com — the Permission Policy primitive

> always_allow vs always_ask defaults per toolset (agent_toolset vs mcp_toolset), per-tool configs[] overrides, and the fact that custom tools are not governed by permission policies at all. Also catalogs invocation_controls, the claude-tag Members/restrict-access primitive — who or what may invoke anything at all, a distinct axis from what a tool may do once invoked.

## API
- GET /api/permission-policies — list all cataloged permission policies
- GET /api/changes — newest-first change feed (?cursor=&limit=) for polling agents
- GET /api/invocation-controls — list invocation controls (the claude-tag Members/restrict-access primitive, generalized), filterable by control_level
- GET /api/invocation-controls/:id — a single invocation control
- POST /mcp — Streamable-HTTP MCP endpoint (initialize / tools/list / tools/call get_permission_policies); server card at /.well-known/mcp/server-card.json

## Grounding
Catalogs the doc's real model: always_allow/always_ask defaults per toolset, per-tool configs[] overrides, and the fact that custom tools are not governed by permission policies at all.
Canonical source: https://platform.claude.com/docs/en/managed-agents/permission-policies.md

## Family
This site is one of the Cloudflare Workers in the subagent*.com family. For the
live, machine-readable directory of every sibling domain, Cloudflare zone ID,
and worker binding, fetch:
https://subagentsubdomains.com/api/dns-zones

For a unified catalog of every API route across every primitive in the family:
https://subagentapi.com/api/catalog