Prompts
Versioned prompt templates with Liquid variables — manage in the dashboard, SDK, or CLI; consume via MCP prompts/list
Prompts
Prompts are reusable instruction templates for your agents. Each prompt supports Liquid variables ({{customer_name}}), version history, and categories. You manage the library through the API, SDK, or CLI. MCP clients consume them as first-class MCP prompts via prompts/list and prompts/get.
MCP exposes three primitives: tools (actions), resources (kb:// documents), and prompts (templated instructions). Workspace prompts from this API are registered automatically on your MCP endpoint.
How it works
- Create a prompt with content and optional variable definitions.
- Version — each content change creates a new version; revert to any prior version.
- Resolve — substitute variables at runtime before sending to an LLM.
- Apply — link a prompt to an agent (
agent.promptId). - MCP — clients call
prompts/listto discover templates andprompts/getwith arguments to render them.
Quick start (CLI)
# List prompts in your workspace
chanl prompts list --status active
# Create from JSON
chanl prompts create -f support-greeting.json
# Update content (creates a new version when content changes)
chanl prompts update <prompt-id> -f updates.json
# Revert to version 2
chanl prompts revert <prompt-id> --to-version 2
# Resolve variables
chanl prompts resolve <prompt-id> -f vars.json
# Attach to an agent
chanl prompts apply <prompt-id> --agent <agent-id>Example support-greeting.json:
{
"name": "Support Greeting",
"content": "Hello {{customer_name}}! How can I help you today?",
"category": "support",
"status": "active",
"variables": [
{
"name": "customer_name",
"type": "string",
"required": true,
"description": "Caller or visitor display name"
}
]
}MCP vs prompt tools
| Surface | What it is | When to use |
|---|---|---|
MCP prompts (prompts/list, prompts/get) | Protocol-native prompt templates from your workspace library | Claude Desktop, Cursor, any MCP client that supports prompts |
Guided prompts (create-agent, review-customer, …) | Static CX workflows shipped with Chanl MCP | Onboarding interviews in MCP clients |
prompt_list / prompt_get tools | Admin tools that call the REST API | Agents that only support tool calling, not MCP prompts |
Manage the library with SDK or CLI. See MCP for connection details.