Bridge AWS Secrets Manager to your AI agents
AWS Secrets Manager is often the source of database passwords, third-party API keys, and rotation-managed application secrets. Rather than issuing every agent AWS credentials, use one dedicated integration role with narrowly defined access to the specific secret ARNs, region, and KMS permissions it requires, then let Gazebo broker mapped values to approved runtimes. Account boundaries, IAM identity policies, resource policies, KMS key policies, and VPC endpoint conditions remain the controls that determine what the AWS role can retrieve.
Why bridge AWS Secrets Manager to AI agents through Gazebo?
Direct per-agent IAM roles can be appropriate for strongly isolated workloads, but they add trust-policy and lifecycle work. A brokered role reduces distribution of AWS credentials while preserving an agent-specific retrieval record. It is not a reason to make the broker role broad: an integration role with secretsmanager:GetSecretValue on '*' or broad KMS decrypt permissions becomes the real blast radius. CloudTrail records AWS API activity, Gazebo records the requesting agent, and downstream logs show use. Disable a profile for fast containment and rotate the secret or revoke role sessions when a delivered value may be compromised.
How it works
- 1
Create a dedicated AWS role with GetSecretValue on explicit secret ARNs, only the required region, and KMS decrypt permission only when the secret’s key requires it. Review the role trust policy and resource policies.
- 2
Map each secret to a named Gazebo profile, separating development and production accounts or secrets rather than relying on a shared label.
- 3
The approved agent requests its credential through MCP; Gazebo evaluates its policy before assuming or using the integration identity to retrieve the value.
- 4
Correlate Gazebo decisions with CloudTrail events, AWS Secrets Manager rotation status, KMS events, and the target service’s logs. Test access-denied, throttling, rotation, and regional endpoint failures.
- 5
Remove a profile to stop future retrieval. Rotate the affected secret, revoke active sessions where applicable, and review CloudTrail if an agent or broker environment may have exposed a credential.
Common use cases
No per-agent IAM roles
A single broker role can avoid placing AWS credentials in every agent runtime, provided its IAM and KMS permissions are limited to explicitly mapped secret ARNs. Gazebo profiles decide which agent may request each value; they do not compensate for an overly broad broker role. Use separate roles or accounts where production isolation requires it.
Audit trail independent of CloudTrail
Gazebo identifies the agent, policy decision, secret mapping, and time; CloudTrail confirms the AWS API operation under the integration role. Combine them for investigations, then consult the destination service for actual credential use. On revocation, remember that neither audit stream invalidates a secret already copied into a running process.
Connect AWS Secrets Manager to Gazebo
Give your agents scoped access to AWS Secrets Manager in minutes. Every call logged. Revoke anytime.
Connect AWS Secrets Manager