Hermes

Branch off your dev environment to create an isolated config so you control the agent's blast radius.

Branch configs are prefixed with the environment slug. dev_agent_hermes lives under dev and inherits from its root unless you override.

Scope it down so the API key for whichever provider you've pointed Hermes at, plus any work credentials the agent actually needs, nothing more. Hermes is multi-provider, so the exact key depends on your hermes model choice.

Do not let the agent populate or modify its own secrets. A human decides what goes in this config. The token in the next step is read-only, so the agent can't write back.

Bound to this one config, with an expiration so a leaked token dies on its own. Capture it into an environment variable so the next steps can reference it without pasting the raw value around.

Service tokens are read-only by default. The agent can fetch secrets, but never change them. This is the agent's identity.

Bind the agent's token to a single directory so the CLI only uses it there. Run this from the folder Hermes will work in.

The --scope . flag locks the token to the current directory. Any doppler command run inside that folder - the doppler run launch below and anything Hermes itself invokes - resolves with this scoped agent token and never your personal CLI token.

Note: Run the agent in a sandbox scoped to that directory. And proceed with caution on production workloads: a non-deterministic agent with live credentials is real blast radius, so keep it pointed at dev/test configs unless you have a specific reason not to.

Hermes resolves credentials from the environment, so doppler run is all that's needed. No config file, and no secrets in your repo. When the process exits, the injected secrets are gone.

Steps 1–6 give Hermes the credentials it runs with. The MCP server lets Hermes see and operate Doppler itself: list projects, inspect a config, and understand the secret layout before depending on it. Useful in almost any session, so set it up alongside the injection above.

The MCP server uses the same scoped, read-only token you minted in step 4. Hermes takes MCP server credentials in a native env: block in config.yaml, so pass the token there and add --read-only because the server can't infer read-only from the token alone.

Alternatively, authenticate once with npx @dopplerhq/mcp-server login and drop the env: block entirely. Verify with hermes mcp list and hermes mcp test doppler.

Note: If you run Hermes alongside other agents, give it its own config so its secrets stay isolated from theirs.