MCP server (AI tools)

The MCP server lets your own AI client — Claude Desktop, Claude Code, Cursor and other Model Context Protocol clients — read and update your Coalface tasks. The AI runs on your side, using your own keys, so there’s no AI cost to you here beyond your own client.

This is different from the AI assist extension, which is Coalface calling AI for you inside a task. MCP is your AI reaching into Coalface.

Turn it on

The MCP server is off by default. Switch it on in Settings → General → Extensions (“AI tools”).

Connect your client

MCP runs on top of the public API, so it uses the same API keys and scopes. In Settings → Integrations you’ll find:

• Your MCP endpoint URL
• A ready-to-paste config snippet for your AI client

Create an API key with the scopes you want the AI to have, drop it into your client’s MCP config alongside the endpoint, and your client can start working with your tasks.

What the AI can do

The MCP tools cover reading and writing the things you’d expect:

• List boards, columns and tasks; get a single task
• Create and update tasks
• Add and list subtasks and notes
• List labels and members

There is deliberately no delete tool — an AI acting on task content it reads shouldn’t be able to destroy anything.

Safe by construction

Because MCP goes through the same layer as the public API:

• Everything is scoped to your workspace and to the scopes on the key.
• Encrypted content is handled the same way, and the Vault is never exposed.
• Creating a task the AI’s way still triggers the normal WordPress sync and webhooks.

Give each key the least it needs — a read-only assistant only needs the :read scopes.

The short version: turn on the MCP extension, mint a scoped API key, and paste the endpoint and key into your AI client. It can then read and update tasks — never delete, never touch the Vault.