GET STARTED
The 5-minute path. Add UPG to Claude Code, Cursor, or any MCP client. Then describe what you’re building in plain language, and it reads and writes the graph as you go.
A real session. Nothing scripted.
The fast path
Not an engineer? You don't have to run anything yourself. Paste the prompt below into Claude Code, Claude Desktop, Cursor, or any MCP-enabled assistant. It installs the CLI, creates your graph file, wires up the MCP server, and adds the /upg-* commands, then verifies the install.
Set up the Unified Product Graph (UPG) on my machine so I can use it with my AI assistant. First read the official setup docs at https://unifiedproductgraph.org/mcp, then walk me through the steps and run them for me.
1. Check my environment: tell me my operating system, and confirm Node.js 18+ and npm are installed. If either is missing, give me the install link for my OS and stop here.
(Windows: if PowerShell blocks npm or npx with an execution-policy error, use the npm.cmd / npx.cmd wrappers, or run the commands in Command Prompt.)
2. Install the UPG CLI globally: npm install -g @unified-product-graph/cli
3. Create my product graph in this folder: upg init --title "My Product" --template blank --yes (the --yes flag is required so it does not stop on an interactive menu; templates: blank, saas, marketplace, mobile, oss; add --force to overwrite)
4. Wire up the MCP server so my AI tool can read and write the graph: upg mcp setup (add --scope user to make it available in every project). That writes the standard .mcp.json entry; the server runs on demand via npx, so there is nothing else to install.
5. Install the /upg-* slash commands: upg install-skills (add --scope user for every project)
6. Verify the install: upg health and upg mcp status (the server should show as connected).
When you are done, tell me where each file was written and how to run my first /upg command. If you hit a permissions error, tell me exactly what access you need.Prefer to run the commands yourself? The steps below walk through the same setup by hand.
Install the CLI once, then run upg mcp setup to wire up any MCP-compatible AI tool. Your graph lives in a local .upg file you control.
npm install -g @unified-product-graph/cliThen run upg mcp setup to wire up your AI tool. Prefer manual MCP config (Cursor, VS Code)?
Brings the full set of /upg-* slash commands into Claude Code: capture, tree, gaps, status, and more.
upg install-skillsDescribe what you're building in plain language. Try prompts like:
Entities connect automatically based on the UPG ontology. Personas link to JTBDs, outcomes link to KPIs, hypotheses link to experiments. The AI handles the structure so you can focus on the thinking.
Start with these six slash commands, the core product-thinking loop. They're part of a larger set of 49 /upg-* commands available inside your AI tool.
/upg-new-graphStart a new product graph and create your first entities/upg-new-personaAdd any entity: personas, hypotheses, features, outcomes/upg-linkLink entities together to build the knowledge graph/upg-show-statusSee your graph dashboard: counts, health, coverage/upg-show-treeVisualize as a tree: OST, OKR, or product hierarchy/upg-check-gapsFind what's missing and get suggestions for next stepsThe full UPG specification: 320+ entity types, 37 domains
Browse every entity type with properties and relationships
42 canonical product frameworks mapped to UPG: RICE, OST, BMC, DORA
How AI tools connect to UPG via the Model Context Protocol
TypeScript types, validation, and constants for tool builders
The case for UPG: essays on why product teams need a graph