Skip to main content

​
Overview

Coconut integrates with the broader AI tooling ecosystem in two complementary ways:
  • Connectors & MCP — a hosted Model Context Protocol server that lets AI assistants (Claude, ChatGPT, Cursor, Windsurf, custom agents) read and write to your Coconut instances using the signed‑in user’s permissions.
  • Skills — a Claude Code plugin marketplace of reusable SKILL.md files that run on a Coconut instance and orchestrate the nut CLI to manipulate tasks, knowledge, context, and connectors.
Together, they cover both how an external AI talks to Coconut (connectors) and how an AI loop running on a Coconut behaves (skills).

Connectors & MCP

Securely extend AI assistants with your Coconut instances using the Model Context Protocol. Hosted at https://app.coconut.dev/mcp, scope‑gated, and OAuth‑authenticated.

Skills

Install reusable Coconut skills via the lovelybunch/coconut-skills Claude Code plugin marketplace, or author new ones with the included meta‑skill.

​
When to use which

  • Use Connectors & MCP when you want an AI assistant outside Coconut (e.g. Claude Desktop, Cursor, ChatGPT) to reach into Coconut on your behalf — listing instances, reading and writing tasks and knowledge, scheduling and triggering jobs, managing skills, and so on.
  • Use Skills when you want to package repeatable behavior that runs inside a Coconut instance — for example, a scheduled workflow that scrapes a news source and writes a summary into .nut/knowledge/.
The two layers are designed to compose: a skill running on a Coconut can be authored, installed, or updated by an external assistant via the MCP connector’s skill_create and skill_update tools.

​
References