Run AI coding agents with Doppler
Give every coding agent its own scoped, read-only, expiring set of credentials. Injected at runtime, kept out of your project files, never editable by the agent itself.
Why this matters
AI coding agents can run commands, call APIs, and edit files on your behalf. To do that they need credentials. The default path is dangerous: drop keys into a .env, let the agent read the whole file, and hope it doesn't echo a secret into a log, a commit, or a prompt it sends upstream.
Doppler helps solve this problem today. There are two halves to connecting an agent, and you'll usually set up both:
- Inject scoped-down secrets at runtime: Use a branch config in Doppler to only hold secrets a given agent needs. You can use a scoped, read-only, expiring service token bound to that config and use doppler run to inject those secrets as environment variables into the agent process at launch. The credentials live in the process environment for the life of the process, never in a .env or project file, and never editable by the agent.
- Connect the Doppler MCP server. The agent gets a read-only window into Doppler itself: listing projects, inspecting configs, understanding the secret layout before it writes code that depends on it.
The result: the agent gets exactly the credentials it needs for the task, plus the context to use them well, and it can't read, modify, or persist anything it shouldn't. Learn how to get started in each of the provider tools below.
Choosing your agent's identity
| Identity | Best for | Notes |
|---|---|---|
Scoped service token (above) | Individual developers, any plan | Read-only, bound to one config, expiring. The default. Simplest path. |
Service Account Identity (OIDC) | Teams, CI, no static tokens | Team plan feature. Cleaner for shared/automated environments, with no long-lived token. Useful when you've outgrown service tokens. |
Why Doppler for agents?
Give your agents only the access they need.
The secrets manager built for the agentic age
Protect secrets across your entire stack with centralized management, automated workflows, and the flexibility to deploy in the cloud or on-prem.
