Get an MCP server's request policy

GET/v1/mcp-servers/{slug}/policy

Get an MCP server's request policy.

What the upstream may ask for while it answers a call: sampling (LLM completions run on a virtual key, with a model, token cap and monthly budget), elicitation (input from the person), roots and the nesting depth, with this month's sampling spend.

Requires an organization admin key, or a key of the server's workspace (any role). Returns 404 when the server does not exist in the organization.

Authorization

AuthorizationBearer va_
A platform API key (va_...), created under Settings > API keys. Scoped to an organization or one workspace, with an admin or member role. Used by the management API.

Path parameters

slugstringrequired
The MCP server's slug (mcps-...).

Query parameters

organizationstringrequired
The organization's identifier (slug).

Responses

200application/json

sampling_enabledboolean
Whether the upstream may ask for LLM completions. Default false. Turning it on needs sampling_virtual_key and sampling_model.
sampling_virtual_keystring
The slug of a virtual key in the server's workspace; sampling completions run, and are billed, on it. It must not be disabled or expired, and must allow sampling_model. Empty clears it.
At least 0 characters · At most 64 characters
sampling_modelstring
The model every sampling request runs on, up to 255 characters; the upstream's model preferences are ignored. Empty clears it.
At least 0 characters · At most 255 characters
sampling_max_tokensinteger<int32>
Most tokens one sampling completion may produce, 1 to 32000. Absent: the gateway default (1024).
Minimum 1 · Maximum 32000
clear_sampling_max_tokensboolean
true in an update removes the server's token cap, back to the gateway default.
sampling_budget_usdnumber
Sampling spend allowed per calendar month (UTC), in USD. Absent: no cap. 0 stops all sampling.
Minimum 0
clear_sampling_budgetboolean
true in an update removes the budget.
sampling_spent_usdnumber
Read-only: what sampling has cost this calendar month (UTC), in USD.
elicitation_enabledboolean
Whether the upstream may ask the person for input. Default true. Requests that ask for a credential are always refused, and guardrails scan each request.
elicitation_url_modeboolean
Whether the upstream may send the person to a web page (an https URL) instead of a form. Default false.
roots_enabledboolean
Whether the upstream may ask which roots the client works in; it is answered with allowed_roots. Default false.
allowed_rootsstring[]
The root URIs the upstream may learn, for example file:///home/agent/project, at most 20. Absolute URIs with no user info or fragment. Replaces the list when sent.
At most 20 items
max_nesting_depthinteger<int32>
How many rounds of sampling, elicitation or roots requests one call may take, 1 to 10. Default 3.
Minimum 1 · Maximum 10

VectorAxis is built by GK Labs, a member of NVIDIA Inception.