MuleSoft AI Gateway metering

MuleSoft AI Gateway and Agent Fabric supply centralized token-metering rows for AI inference traffic routed through MuleSoft.

Release state

This guide documents the read-only mulesoft connector implemented in the platform connector surface.

Required access

The names below are Venturi manifest capabilities, not Anypoint scopes:

Venturi capability	Purpose
AnypointPlatform.Read	Read Anypoint platform context.
AIGateway.Metering.Read	Read AI Gateway token metering rows.
AgentFabric.Read	Read Agent Fabric context.

Do not grant API deployment, policy mutation, or gateway administration scopes.

Use an Anypoint connected app and assign only organization and environment roles that can read the configured metering and inventory resources. See MuleSoft’s connected-app documentation. The connector never submits the Venturi capability names above as OAuth scopes.

Setup

1. Create a dedicated Anypoint integration identity.
2. Grant only the read-only scopes above.
3. Store the token reference in Venturi.
4. Set VENTURI_MULESOFT_LOOKBACK_DAYS if you need a poll window other than 30 days.
5. In Venturi, open Integrations and select MuleSoft AI Gateway and run Test connection.

Verification

• Gateway metering rows appear for the expected gateway identifiers.
• Token counts are normalized with source mulesoft_ai_gateway_metering_api.
• Upstream read failures return no rows rather than interrupting customer traffic.
• The read-only inventory test succeeds without API deployment or policy-write access.

Rotation and offboarding

Rotate the Anypoint credential through MuleSoft administration. Removing the connector stops new MuleSoft AI Gateway metering syncs.