Skip to main content

​
Why connectors and MCP matter

The Model Context Protocol (MCP) is an open standard that lets AI assistants — like Claude, ChatGPT, Cursor, and others — securely talk to external systems through a common interface. A connector is simply an MCP server that exposes a specific product or service (in our case, Coconut) as a set of tools the AI can call on your behalf. Instead of copy‑pasting between tools or building bespoke integrations, connectors give your AI assistant real, live access to the systems where your work actually lives, while preserving the permissions of the signed‑in user. If you can’t see something in the source system, the connector can’t reach it either. Key benefits:
  • Live context — the assistant reads current state (tasks, jobs, knowledge) instead of stale snapshots.
  • Actionable — beyond reading, it can create and update records on your behalf.
  • Permission‑aware — every request runs as you, subject to your Coconut scopes and role.
  • Portable — any MCP‑compatible client works the same way (Claude, Cursor, Windsurf, ChatGPT, custom agents, etc.).

​
The Coconut connector

The Coconut connector is an MCP server hosted at:
Once connected, your AI assistant can reach across every Coconut instance you have access to — listing instances, inspecting their health, reading and writing tasks and knowledge, pulling context, and browsing jobs — all with per‑scope authorization.

​
What it can access

The Coconut MCP server exposes the following tools. Each tool is gated by a specific OAuth scope, so you can grant read‑only access or allow writes on a per‑capability basis.
To call these Coconut MCP tools from the CLI, use nut coconut. Prefer typed wrappers like nut coconut task list --name <name>, nut coconut skill list --name <name>, and nut coconut job list --name <name> when available. Use nut coconut call <tool_name> as the raw escape hatch for newly added tools.
Scopes are requested during the initial OAuth flow. You can re‑authenticate at any time to broaden or narrow the tools your assistant can invoke.
task_review and task_start will not start a new agent session if the task status is already active. In that case the tool returns a message indicating the task is already active and no new session was started.
skills_create, skills_update, and skills_delete are admin-only write capabilities for skill management. They are only exposed to Clerk organization admins and only apply to org-owned coconuts in the user’s active organization context.

​
Add the Coconut connector to Claude

Claude supports remote MCP servers as custom connectors. The steps below are adapted from Anthropic’s official guide.
Custom connectors are available on Claude, Cowork, and Claude Desktop for users on Free, Pro, Max, Team, and Enterprise plans. Free users are limited to one custom connector. On Team and Enterprise plans, an Owner or Primary Owner must enable the connector for the organization first.
1

Open connector settings

In Claude, navigate to Customize → Connectors. Click the + button next to Connectors.
2

Add a custom connector

Select Add custom connector.
3

Enter the Coconut MCP details

Use the following values:
  • Name: Coconut
  • URL: https://app.coconut.dev/mcp
Leave advanced settings (OAuth Client ID and secret) empty unless your organization requires them.
4

Connect and authenticate

Click Add, then follow the OAuth prompts to sign in to Coconut and grant the scopes you want the assistant to use (for example, tasks:read, knowledge:write).
5

Enable it in a conversation

In any chat, click the + button (or type /), hover over Connectors, and toggle Coconut on. Claude can now call Coconut tools whenever it’s useful — e.g. “List open tasks on my main Coconut” or “Create a task titled ‘Audit login flow’ on coconut acme-prod.”
You can also open the ready‑made setup URL directly: https://app.coconut.dev/connectors/coconut.

​
Other MCP clients

Because the Coconut connector is a standard remote MCP server, any MCP‑capable client works the same way. Point your client at:
…and complete the OAuth flow. See our guides for Cursor, Windsurf, and Claude Code for client‑specific setup details.

​
Security and permissions

  • All traffic to https://app.coconut.dev/mcp is encrypted in transit.
  • The connector never escalates privileges — tool calls execute as the authenticated user, subject to your Coconut role and scopes.
  • You can disconnect the connector at any time from your AI client’s connector settings, which revokes its ability to call Coconut tools.
  • On Team and Enterprise Claude plans, Owners can restrict write‑scoped tools (e.g. allow tasks:read while blocking tasks:write) org‑wide.

​
References