Credentials, Secrets, And Accounts
Apprentice connects to providers, integrations, and MCP servers through credentials you configure.
Treat all provider keys, OAuth accounts, integration tokens, webhook URLs, and MCP secrets as sensitive.
Credential Surfaces
Credentials can appear in:
- AI provider settings.
- Provider accounts.
- External Communications integrations.
- MCP Library secrets.
- Per-agent MCP secret overrides.
- Webhook authorization headers.
- Local runtime API keys, if enabled.
Do not paste credentials into agent prompts, chats, memory, knowledge files, or docs.
Provider Accounts
Provider accounts authenticate model access.
Use separate accounts or keys when:
- You want clean billing boundaries.
- Test agents should not use production credentials.
- Different agents need different provider accounts.
- You want easy revocation for one workflow.
If an agent's provider account changes, provider conversation continuity can be retired so future runs do not mix account context.
Integration Credentials
Integrations are configured globally in Settings > External Communications.
Credentials are stored through the available credential store. If the app reports a credential issue, edit the integration and update credentials before relying on the channel.
Use dedicated service accounts when possible.
For WhatsApp, use a dedicated number when possible because linked-device automation carries account risk.
MCP Secrets
MCP server secrets are configured globally in the MCP Library.
Agent bindings can use the global default secret or override a secret for that specific agent.
Use per-agent overrides for:
- Read-only tokens.
- Test versus production environments.
- Separate customer or project scopes.
- Least-privilege service accounts.
Webhooks
Webhook URLs and Authorization headers are credentials.
Keep separate endpoints for testing and production. Rotate webhook secrets if they are exposed.
Practical Safety Checklist
Before enabling an agent:
- Confirm credentials are stored in settings, not prompts.
- Use the least-privilege account available.
- Use a dedicated service account when practical.
- Avoid sharing one production token across unrelated agents.
- Review per-agent MCP secret overrides.
- Test with a low-risk prompt first.
Troubleshooting
If a provider fails, re-test the provider account in AI Integration.
If an integration reports a credential issue, update it in External Communications.
If MCP tools fail due to auth, check MCP Library secrets and per-agent overrides.
If a credential was exposed, rotate it in the external service and update Apprentice.
Next Step
After credentials are safe, review permissions and guardrails for the agent that will use them.