Skip to content
Coro Docs

Claude Code authentication

The built-in @coro-ai/llm-anthropic executor reads auth from plugins.installed.anthropic.config, which you configure in the dashboard—not by hand-editing secrets into markdown.

  1. Open Settings → Setup → LLM provider.
  2. In the Anthropic panel, choose an auth method and complete the flow.

LLM provider settings

Method in UImethod valueBest for …
Claude loginclaudeLoginRecommended for laptops — Claude Code persists an interactive OAuth-style session locally; click Connect with Claude and follow the browser prompt.
API keyapiKeyCI bots or regulated environments where unattended keys are audited centrally.
Legacy OAuth tokenoauthMigration from claude setup-token only — avoid for net-new setups.

Click Save all changes, then Test connection before starting a run.

Why Claude login wins for humans

  • No shared secrets in plaintext chat logs — each developer authenticates individually.
  • Automatic refresh aligns with Claude Code lifecycle instead of brittle copy/paste PAT rotation.
  • Org policy alignment lets enterprise bans on key-based access stick while Coro stays usable.

API key deployments

When automation forbids OAuth devices:

  1. Provision a restricted Anthropic (or routed Bedrock/GCP) credential per sandbox.
  2. Enter it under Settings → LLM provider → API key (or inject via your install automation into plugins.installed.anthropic.config).
  3. Lock down ~/.coro/config.json permissions on shared jump boxes — Settings → Advanced → Paths shows the config file location.

Combine with budget alerts downstream because keys frequently lack per-seat usage visibility.

Switching modes safely

Flipping auth method mid-flight does not cancel active jobs — finish runs, then restart coro start after saving if auth pivots behave unexpectedly (/guides/hybrid-mode/ team runners).