Developers Get an API key
Platform

Audit logs

What Sentinel records about API calls, MCP tool calls and connections, so a firm can answer "what was read, and on whose behalf?"

REST API

These events go to the firm's audit log, which admins see under Settings → Audit log. Each is stamped as an API-key call, with the key's ID.

Event Recorded when Details
api_v1_search A search runs The matter, the mode, the query's length and the number of results. The query text itself isn't stored.
api_v1_document_read A document is read with GET /documents/{id} The document, its matter, and whether the text was included

MCP tool calls

Every MCP tool call writes one record, whether it succeeds or fails:

Field Description
user_id The Sentinel user the call ran as
api_key_id The connection or API key used
tool_name For example, search_documents
args The arguments the client sent, such as the matter, query and filters
result_size_bytes Size of the result returned
duration_ms How long the call took
error The error message, if the call failed
created_at When the call happened
Note

The MCP tool-call log isn't shown in the app yet. Firm admins can ask SNT Labs support for an export for a user, matter or date range.

Connections and sign-ins

Event Recorded when
oauth_authorized A user signs in and approves an app
oauth_token_issued An app exchanges its code and a connection key is created
oauth_authorize_failed A sign-in is denied, fails, or comes from an account that can't connect
oauth_token_refused A token request is rejected
api_key_rejected A request uses an invalid, revoked or expired key

Failure events never store the submitted password or key. At most they keep the first 12 characters of a well-formed key (its display prefix, such as sk_live_8f14), so the firm can tell which key was used.