Inspect tool-call evidence

Tool-call evidence answers whether Axec admitted a capability, which safe authority facts were involved, and what bounded upstream outcome followed. It is not a request-body archive or a source of credentials.

Prerequisites

• The target Application ID and tenant key.
• An active administrator session or a member session allowed to view that User’s own events.
• A correlation ID or approximate call time from the client.

Inspect the timeline

In the Console, open Applications, select the gateway Application, and open Tool calls. Or read the cursor-paginated projection:

GET /api/v1/ws/{key}/applications/{application_id}/tool-calls

Filter by decision or outcome when narrowing a large timeline. Follow next_before_at and next_before_id together; they form one composite cursor.

For the target correlation ID, check:

1. mcp.tool.decision, the authoritative allow or deny result;
2. the principal kind and safe User or Application/workload attribution;
3. the gateway, Connector, and capability identifiers;
4. the decision reason and evaluator version;
5. the optional correlated upstream.outcome; and
6. any approval or Data Protection state associated with the same governed action.

The fixed authority evaluator currently reports its decision, reason, and evaluator version rather than inventing a named policy attribution.

Interpret missing outcomes

• A deny has no upstream outcome because Axec rejected it before dispatch.
• awaiting_approval requires approval and an authenticated identical retry; polling alone never executes the action.
• outcome_unknown means dispatch may have happened but the result is not safely known; do not automatically replay a mutation.
• data_protection_unavailable means Axec withheld a result it could not safely transform, not that the raw result is available in evidence.

Evidence boundary

The projection must not contain access or refresh tokens, client secrets, encrypted credential values, Authorization or Cookie headers, raw arguments, unrestricted request bodies, unsanitized results, or detected sensitive values. ActionRequest-backed results retain only the complete sanitized MCP envelope or sanitized proxy response for the bounded retention period.

For the full enforcement order, see Evidence and Security Architecture.