Connect Doppler secrets to your AI agents via Gazebo
Doppler organizes secrets by project and configuration, which is useful for development, staging, and production but can make a service token far broader than an individual agent task. Keep configuration boundaries meaningful, create a machine-oriented token limited to the project/configuration Gazebo needs, and map only the required secret names to agent profiles. This lets an agent obtain a runtime value without receiving direct Doppler credentials or a general ability to enumerate a configuration.
Why route Doppler secrets through Gazebo for agents?
Doppler’s service-token scope is the upstream capability boundary; Gazebo does not reduce what a stolen upstream token could read. It does prevent normal agents from receiving that token and applies a separate, secret-level retrieval policy. Use separate configs for materially different environments, correlate Gazebo access decisions with Doppler activity and downstream service logs, and plan for the fact that removing a profile stops future retrieval but cannot retract a value already delivered to a process.
How it works
- 1
Create a Doppler service token with the narrowest relevant project and configuration access; keep production and non-production tokens distinct.
- 2
Map individual secret names to named Gazebo profiles instead of handing an agent a Doppler token or a complete environment.
- 3
The agent requests its credential through MCP and Gazebo checks the profile before making the upstream retrieval.
- 4
Review Gazebo’s allowed and denied requests with Doppler events and the downstream provider’s activity, especially after retries or deployment failures.
- 5
Disable the profile for immediate containment; rotate the Doppler token or affected secret upstream when exposure is plausible, then update the authorized mapping.
Common use cases
Dev and prod environment separation
Use separate Doppler configurations and service tokens for development, staging, and production, then map those values to separate Gazebo profiles. This makes an environment-targeting mistake visible in both configuration and policy, and allows a production token to be rotated without interrupting development automation.
Secret-level access control
Gazebo can limit routine agent retrieval to SECRET_A rather than SECRET_B, even when its upstream token can read the configuration. Keep that upstream token narrowly scoped and treat it as sensitive infrastructure. If it is exposed, rotate it: deleting an agent profile alone does not protect the rest of the configuration.
Connect Doppler to Gazebo
Give your agents scoped access to Doppler in minutes. Every call logged. Revoke anytime.
Connect Doppler