Product Feature Graph

1. 1

   Create a workspace

   Go to My workspaces and click New workspace. A personal workspace is also auto-created on first login. Each workspace holds one or more product graphs.

2. 2

   Generate a Personal Access Token (PAT)

   Go to Account settings → Tokens and create a PAT. Tokens are user-scoped and work across all your workspaces. Copy it immediately — it's only shown once. HMAC-SHA256 hashed at rest.

3. 3

   Connect your agent via MCP

   Add the MCP server config to your agent (Claude Code, Cursor, VS Code Copilot, etc.):

   {
     "mcpServers": {
       "product-graph": {
         "type": "http",
         "url": "https://graphapi.vibelevel.ai/mcp?workspace=<your-workspace>",
         "headers": { "Authorization": "Bearer <your-pat>" }
       }
     }
   }

   Compatible with Claude Code · Cursor · Windsurf · VS Code Copilot · any MCP HTTP client.

4. 4

   Seed your graph — run from your product repo

   Open your product repo in your coding agent (with the MCP server connected above). The agent needs to be inside the codebase to scan it properly. Then ask:

   "I'm in the [product-name] codebase. Scan this project and create a product graph for workspace <your-workspace>. Identify the main product, its key features, any reusable agent skills or patterns, and standards the project follows. Namespace product-specific node ids under the product (e.g. [product].[feature]); use bare ids only for shared nodes."

   The agent will scan your code, propose nodes and edges, and submit them for your review.

5. 5

   Keep it updated — at the end of each coding session

   After adding features, enhancements, or new patterns in a session, ask your agent to sync:

   "Based on the changes we implemented in this session, update the product graph in workspace <your-workspace>. Add any new features, update existing ones, and if we introduced any reusable patterns, standards, or agent skills, add those too."

   Run this from inside your repo so the agent has full context of what changed.

6. 6

   Review and approve proposals

   Agent-proposed changes land in the Pending tab of your workspace admin. Each proposal shows a structured diff — new and updated nodes grouped by type. Review, then one-click approve or reject. Only approved changes become live.

A workspace is your team's shared graph. It can hold several products, and that's the point: agent skills, standards, and personas defined once in the workspace apply across all of its products — define a coding standard or reusable skill and every product (and every agent) inherits it via MCP. Cross-product relationships (e.g. one product integrates with another) live in the same graph. Each workspace has a unique slug, its own members and settings, and can be public or private.

Common setups

A product graph is a structured knowledge map of your product — its features, personas, agent skills, standards, and the relationships between them. It's readable by both humans and AI agents via MCP. Schema version: v1.8 (13 node types, 14 edge relationships).

Node types

Edge relationships

The interactive graph viewer (vis-network) shows each item type in a distinct color and symbol. Open it from any workspace via Graph in the nav. The left panel controls which items appear; the Display panel (top-right) hides categories. The four main controls are numbered ①–④ to match the in-app Graph Viewer Guide.

Accessible at /p/[workspace]/admin. Workspace members see a sidebar with Home, Pending, Graph, Audit, and Settings.

The MCP HTTP transport is the agent interface — 12 tools: 10 read + 2 write (propose_graph_diff, set_workspace_settings). A valid PAT is required for all access. Usage is tracked per-token per-day.

{
  "mcpServers": {
    "product-graph": {
      "type": "http",
      "url": "https://graphapi.vibelevel.ai/mcp?workspace=<your-workspace>",
      "headers": { "Authorization": "Bearer <your-pat>" }
    }
  }
}

Run all of these from inside your product repo so your agent has full codebase context. Use with Claude Code, Cursor, VS Code Copilot, or any MCP-connected agent.

"I'm in the [product-name] codebase. Scan this project and create a product graph for workspace [your-workspace]. Identify the main product, its key features, any reusable agent skills or patterns, and standards the project follows. Namespace product-specific node ids under the product (e.g. [product].[feature]); use bare ids only for shared nodes."

"Based on the changes we implemented in this session, update the product graph in workspace [your-workspace]. Add any new features, update existing ones, and if we introduced any reusable patterns, standards, or agent skills, add those too."

"I just added [feature name] to [product-name]. Update the product graph in workspace [your-workspace] to include this new feature, its relationships to existing nodes, and any new agent skills or standards it introduces."

"We've established a consistent pattern for [pattern name] in this codebase. Add it as a reusable agent_skill node to workspace [your-workspace] with instructions so any agent working on this repo can follow the same pattern."

"Add a standard node to workspace [your-workspace] for [standard name] — this is a rule all agents working on this codebase must follow. Include rationale, examples of correct usage, and anti-examples."

"Before we start, use the product graph for workspace [your-workspace] to understand the existing features and agent skills. Then implement [new feature] consistently with the patterns already in the graph."