ai-cred

Centralized credential store for AI sessions. Store SSH keys, Jenkins tokens, Portainer logins, AWS credentials, and API keys in the macOS Keychain — Claude fetches them on demand via MCP. No more re-pasting secrets every session.

Install

npm install -g ai-cred

Setup with Claude Code

Register the MCP server so Claude can access your credentials:

claude mcp add ai-cred -- ai-cred-server

Then in your CLAUDE.md, add:

## Credential Store (ai-cred MCP)
- Use `get_credential` to fetch creds before connecting to any service
- Use `find_credential` with a partial name if unsure of the exact service name
- Use `list_credentials` to see what's available
- Use `store_credential` to save new credentials
- Use `delete_credential` to remove credentials

CLI Usage

# Add a credential (interactive, masked input)
ai-cred add my-server -t ssh -e prod

# List all credentials (masked preview, no full secrets)
ai-cred list

# Get full credential details
ai-cred get my-server -e prod

# Update a credential
ai-cred update my-server -t ssh -e prod --host 10.0.0.1

# Remove a credential
ai-cred remove my-server -e prod

MCP Tools

Once registered, Claude has access to these tools:

| Tool | Description | |------|-------------| | store_credential | Store a credential in the macOS Keychain | | get_credential | Retrieve a credential by service name and environment | | list_credentials | List all credentials (metadata only, no secrets) | | delete_credential | Remove a credential | | find_credential | Fuzzy search by partial service name |

Credential Types

| Type | Fields | |------|--------| | ssh | host, port, username, password/keyPath | | jenkins | url, username, apiToken | | portainer | url, apiToken | | aws | accessKeyId, secretAccessKey, region, profile | | api-key | url, key |

All types support optional notes field.

Environments

Credentials are organized by environment: prod, dev, staging, global

Security

• Credentials stored in a dedicated macOS Keychain (ai-cred.keychain-db) — isolated from your login keychain
• All Keychain operations use parameterized subprocess calls — no shell injection possible
• List operations never expose secrets — metadata only
• Audit log at ~/.ai-cred/audit.log records every access (timestamps and operations only, never credential values)
• Error messages are sanitized — no credential data in logs or stack traces

Requirements

• macOS (uses macOS Keychain via security CLI)
• Node.js 22+

License

ISC