Cursor

Agent skills + MCP server integration

Cursor supports both agent skills (via .cursor/skills/) and MCP servers. We recommend starting with the skill so the agent has full context on workflows, runs, and collections; add MCP if you want Cognetivy tools to appear as chat tools. You can use both.

Prerequisites

Node.js 18+ and a project directory (or empty folder) where you want the .cognetivy/ workspace. No global install required for the skill-only path; for MCP, Cognetivy must be on your PATH (global install or ensure cognetivy resolves in the shell Cursor uses).

Method 1 - Agent skills (recommended)

Run the installer from your project directory. It will create the workspace, install the skill, and open Studio.

npx cognetivy

Choose Cursor in the installer. Cognetivy will:

Create a .cognetivy/ workspace in your project
Install the Cognetivy skill into .cursor/skills/cognetivy/
Optionally apply a workflow template
Open Studio in your browser

To install or update the skill only (workspace already exists and Cognetivy is installed globally):

cognetivy install cursor

This writes SKILL.md and REFERENCE.md into .cursor/skills/cognetivy/.

Method 2 - MCP server

Connecting via MCP exposes Cognetivy tools (workflow_get, run_start, run_step, collection_append, etc.) directly in Cursor’s chat. The agent can call them as tools without relying solely on the skill text.

Via Cursor settings UI

Open Settings → Tools & MCP (Mac: Cmd+Shift+J, Windows/Linux: Ctrl+Shift+J). Restart Cursor after changing MCP config.

1Open Settings → Tools & MCP (or Features → MCP)
2Click Add new MCP server
3Set Name to cognetivy
4Set Command to cognetivy
5Set Arguments to mcp
6Save and restart Cursor

Via config file

Add the server to ~/.cursor/mcp.json or your project’s .cursor/mcp.json:

{
  "mcpServers": {
    "cognetivy": {
      "command": "cognetivy",
      "args": ["mcp"]
    }
  }
}

Workspace not in project root?

If .cognetivy/ is in a different path, pass it explicitly:

"args": ["mcp", "--workspace", "/path/to/.cognetivy"]

What the agent can do

With the skill or MCP, the agent can run the full CLI surface: workflow list/get/set, run start/status/step/complete, event append, collection-schema get/set, collection list/get/set/append, node-result list/get/set. Key commands:

cognetivy workflow list
cognetivy workflow create --name <string> [--id <string>] [--description <string>]
cognetivy workflow select --workflow <id>
cognetivy workflow get [--workflow <id>] [--version <id>]
cognetivy run start --input <path> [--name <string>]
cognetivy run status --run <id> [--json]
cognetivy run step --run <id> [--node <id>] [--collection-kind <kind>]
cognetivy run complete --run <id>

Full list: CLI reference and MCP tools.

Create your first workflow Full docs on GitHub