DuploCloud Plugin for Claude Code
Ticket-driven workflow enforcement for Claude Code. Integrates with DuploCloud AI Service Desk to ensure all code changes are tied to a ticket, spec, and plan.
Prerequisites
- Claude Code installed
- Access to a DuploCloud instance with AI Service Desk enabled
Installation
Clone this repo and run the install script:
git clone [email protected]:duplocloud/duplo-claude-plugin.git ~/.claude/plugins/duplo
~/.claude/plugins/duplo/scripts/install-bin.sh
The install script:
- Installs CLI tools to
~/.duplocloud/bin/
- Adds them to Claude Code's pre-approved tools
- Adds a shell alias so
claude always loads this plugin
Restart your shell (or source ~/.zshrc), then use claude as normal.
Getting Started
- Activate your project — run
/duplo:activate_project in Claude Code. This logs you in and sets the project context.
- Activate a ticket — run
/duplo:activate_ticket to create or resume a ticket.
- Write a spec — run
/duplo:write_spec to draft and save a specification.
- Write a plan — run
/duplo:write_plan to draft and save an implementation plan.
- Deactivate — run
/duplo:deactivate to clear credentials and project context.
Upgrading
Run the upgrade script to pull the latest version:
~/.claude/plugins/duplo/scripts/upgrade.sh
Uninstall
rm -rf ~/.claude/plugins/duplo
rm -rf ~/.duplocloud
Also remove the plugin alias from your ~/.zshrc / ~/.bashrc (the line containing plugin-dir.*duplo).
Environment Variables
Copy .env.example to .env and fill in your values. Run source .env before launching Claude.
| Variable | Required | Description |
|---|
DUPLO_TOKEN | ✅ Always | Bearer token for the DuploCloud AI helpdesk API. |
DUPLO_HELPDESK_URL | ✅ Always | Base URL of the duplo-helpdesk MCP server (e.g. https://mcp-ai-studio.your-company.duplocloud.net). |
DUPLO_AGENT_MODE | Optional | Controls who responds to ticket messages. See below. |
DUPLO_AGENT_MODE
| Value | Behaviour |
|---|
true (default) | Remote agent mode. The DuploCloud backend AI agent handles all ticket responses via Ticket_send_message_streaming. The backend enriches every request with workspace credentials, scopes, personas, and secrets before the agent runs. Use this for production workflows. |
false or unset | Local agent mode. Claude / Cursor itself acts as the agent. Messages are recorded to the ticket via message_mode=1 (record-only — no backend AI invoked). Claude fetches workspace context (personas, scopes, project spec/plan) at activation and caches it locally with a 5-minute TTL. Credentials come from the developer's own machine (~/.aws/credentials, ~/.kube/config, etc.). Use this when you want the AI coding assistant to drive the conversation directly. |
Which mode should I use?
- Most users: leave
DUPLO_AGENT_MODE=true. The remote agent has full scope credentials and runs in a managed environment.
- Plugin developers / power users: set
DUPLO_AGENT_MODE=false to use Claude/Cursor as the agent with your own machine credentials.
