@onebtn/cli

OneFlow Pro AI Assistant for Claude Code, Cursor, Windsurf, and VS Code — an MCP server that lets your AI coding agent manage products, orders, customers, pages, and automations on the OneFlow Pro platform using plain business language.

Install

npm install -g @onebtn/cli

Requires Node.js ≥ 18.

Quick start

ob-init                # scaffold a workspace at ~/ob (or pass a path)
cd ~/ob
# open the workspace in your AI tool of choice

ob-init creates:

| File | Purpose | |---|---| | CLAUDE.md | Instructions for Claude Code | | .cursorrules | Instructions for Cursor | | .windsurfrules | Instructions for Windsurf | | .github/copilot-instructions.md | Instructions for VS Code Copilot | | .claude/settings.json, .vscode/mcp.json | MCP server registration | | .ob/config.json | OneFlow Pro server URLs |

The four AI-prompt files start from a bundled snapshot and auto-refresh from the OneFlow Pro server on every session (Claude Code) or via ob-cli refresh (other editors).

Using with Claude Desktop

Claude Desktop (the chat app, including its Claude Code mode) doesn't read project-level .mcp.json — it loads MCP servers from a central config file. Add ob there once:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json Linux: ~/.config/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "ob": {
      "command": "npx",
      "args": ["-y", "-p", "@onebtn/cli", "ob-cli"]
    }
  }
}

Fully quit Claude Desktop (⌘Q on Mac, not just close the window) and relaunch.

Standalone Claude Code (CLI) reads .mcp.json from the workspace, so no extra step is needed there — ob-init already wires it up.

First session

In your AI tool, ask the agent to run ob_whoami. It will:

1. Show your CLI version and session state.
2. Prompt for login if you're not authenticated.
3. Pull the latest AI assistant instructions from the server and inject them into the conversation.

After logging in (ob_login), select an organization (ob_select_org) — this loads the entity-type catalog so the assistant can work with your data model.

MCP tools

The package exposes five tools to MCP-aware clients:

| Tool | Purpose | |---|---| | ob_whoami | Session info + authoritative AI instructions | | ob_login | Authenticate with email + password | | ob_select_org | Pick an organization to work in | | ob_execute_steps | Run OBP automation DSL steps (read/write/query/loops/conditions) | | ob_switch_env | Switch between demo and prod environments |

CLI commands

ob-cli              # start the MCP server (used by your AI client; you rarely run this directly)
ob-cli refresh      # re-fetch the AI prompt from the server and rewrite local prompt files
ob-cli help         # show help
ob-init [dir]       # scaffold a new workspace (default: ~/ob)

ob-cli refresh is for users who don't use Claude Code — Cursor, Windsurf, and Copilot read the prompt from static files, so a refresh updates them in one shot.

Configuration

The workspace's .ob/config.json controls which OneFlow Pro server you talk to:

{
  "servers": {
    "prod": "https://oneflowpro.shaps.co.il",
    "demo": "https://devserver.tailae99dc.ts.net"
  },
  "current_env": "demo"
}

Switch in-session with ob_switch_env.

How prompt updates work

The AI assistant's operating instructions live on the OneFlow Pro server. Every ob_whoami call fetches them (with 5-minute cache + ETag) and:

• Returns them inline so Claude Code uses them directly.
• Writes them to CLAUDE.md / .cursorrules / .windsurfrules / .github/copilot-instructions.md so non-MCP-aware editors stay current too.

So when the platform team updates how the assistant should behave, every installed CLI picks it up on the next session — no npm update, no re-init.

License

MIT