Set up a coding agent
Setting up an agent takes four steps: install the CLI where the agent executes commands, give it a token, pin the environment, and add an instruction block.
Install the CLI where the agent runs
The CLI must be on the PATH of the shell the agent uses. That shell is not always your own machine.
To check the installation, run:
Provide a token
Create a personal access token, if not already created; see Quickstart for how. Scope it to only what the agent’s tasks require.
Use the pattern from Tokens for agents, CI, and scripts: the OS keychain on a developer machine, an injected secret in containers or CI.
The agent runs deel commands; the CLI resolves the token, and the agent is not exposed to the token value.
Pin the environment
Agents explore by running commands. Keep them in the demo environment until a workflow is proven.
When the workflow is ready for production, set DEEL_ENV=prod for that agent explicitly, or instruct the agent to pass --env prod on the commands that need it.
Add the instruction block
Agents follow project instruction files. Add the block below to the file your agent reads.
Claude Code
Cursor
Codex
Other agents
Append to CLAUDE.md in the repository root (or ~/.claude/CLAUDE.md for every project).
Verify
Ask the agent to confirm its setup before giving it real work, using the same two checks from Quickstart:
- “Run
deel auth status --fields valid,sourceand tell me the result.” - “Describe the request body of
deel adjustments create --invoiceusing its--help --jsonoutput.”
Checklist for hosted and shared environments
Beyond the general guardrails, hosted and shared environments also require:
- Use the installer in containers; the binary needs no runtime.
- Set
DEEL_TOKENfrom the platform’s secret store, never from a file in the workspace.