Connection policy

The connection family is the widest governed lane an agent reaches. These three reads report what policy actually says about it.

On every AccountKit this build can create, the answer is mode: "open" — every toolkit and every tool in it is reachable.Both kit-creation paths hardcode mode: "open", which omits the toolkit list entirely. The engine implements per-toolkit and per-tool filtering in full; it is inert because no kit ever names a toolkit. wide: true is a field of its own precisely so this is not something you have to infer.

The resolved policy

GET /v1/connections/policy?tenant=<tenant_user_id>

{
  "tenant_user_id": "8f1c…",
  "kit_id": "kit-…",
  "mode": "open",
  "wide": true,
  "wide_because": "mode 'open' omits the toolkit list entirely, so every Composio toolkit and every tool in it is reachable. The engine implements toolkitAllowed/toolAllowed in full; nothing is calling it.",
  "toolkits": null,
  "tools": {},
  "approval": {
    "requires_approval": null,
    "approval_toolkits": null,
    "connect_action_id": "connections.connect",
    "connect_defaults_to_approval": true
  },
  "custom_auth_configs": []
}

`toolkits: null` does not mean “none”

Under mode: "open" the toolkit list is omitted, not empty. null here means all of them. An empty array would read as the opposite, which is why the API returns null and why wide exists as its own boolean.

`mode`	`toolkits`	Meaning
`open`	`null`	every toolkit is reachable
`allowlist`	array	only the listed toolkits
`denylist`	array	every toolkit except the listed ones

Connecting versus executing

Two different decisions, and only the first has an action id.

Connecting is connections.connect, a regular action id that defaults to human approval (connect_defaults_to_approval).
Executing a tool through a connection is a different evaluation kind entirely, dispatched with the toolkit and tool as data. There is no connections.execute:<toolkit> action id, and minting one would create a third representation of a decision the engine already makes two ways.

Which toolkits policy names

GET /v1/connections/toolkits?tenant=<tenant_user_id>

This answers from the policy, never from the provider catalogue. Under mode: "open" policy names nothing, so items is null and governed_count is 0 — the honest answer to “which toolkits does policy name”.This is deliberately not a listing of the provider’s ~1,000 toolkits. That listing is already reachable elsewhere; a second copy of it under a policy-adjacent path would read as an endorsement.

200

{
  "mode": "open",
  "wide": true,
  "wide_because": "…",
  "items": null,
  "governed_count": 0,
  "note": "this list is what POLICY names, not what the provider offers. Under mode 'open' policy names nothing and every toolkit is reachable."
}

The per-toolkit tool filter

GET /v1/connections/toolkits/{toolkit}/tools?tenant=<tenant_user_id>

200

{
  "toolkit": "gmail",
  "mode": "open",
  "wide": true,
  "toolkit_reachable": true,
  "enable": null,
  "disable": null,
  "approval_required": false,
  "note": "enable/disable are the per-toolkit tool filter the engine enforces. null means the policy sets no filter for this toolkit, which under mode 'open' means every tool in it is reachable."
}

Field	Meaning
`toolkit_reachable`	Under `open`, always `true`. Under `allowlist`, whether the toolkit is listed. Under `denylist`, whether it is not.
`enable` / `disable`	The tool filter for this toolkit. `null` means no filter is set — which under `open` means every tool is reachable.
`approval_required`	Whether executing against this toolkit requires human approval for this tenant.

Connections API — the tenant-facing connection surface
GET /v1/limits — reports the same connections fact in surface_honesty

{ "tenant_user_id": "8f1c…", "kit_id": "kit-…", "mode": "open", "wide": true, "wide_because": "mode 'open' omits the toolkit list entirely, so every Composio toolkit and every tool in it is reachable. The engine implements toolkitAllowed/toolAllowed in full; nothing is calling it.", "toolkits": null, "tools": {}, "approval": { "requires_approval": null, "approval_toolkits": null, "connect_action_id": "connections.connect", "connect_defaults_to_approval": true }, "custom_auth_configs": [] }

​The resolved policy

​toolkits: null does not mean “none”

​Connecting versus executing

​Which toolkits policy names

​The per-toolkit tool filter

​Related

The resolved policy

`toolkits: null` does not mean “none”

Connecting versus executing

Which toolkits policy names

The per-toolkit tool filter

Related