Plugin, MCP and skills

Two ways to give your AI agent Mark, the plugin or an MCP connection plus skills, and which one to pick if you work in more than one Mark workspace.

Your agent gets Mark from two parts:

The MCP connection. It gives the agent Mark's tools, signed in as you, in one workspace.
The skills. A skill is a short playbook the agent follows for one job, such as planning a week of posts. Skills call the same tools as any other request, so they do not bypass review.

The Mark plugin is those two parts in one install. You can also add them separately. Both come from the same source: github.com/Life-With-Data/mark-agent-plugin. The skills are written alongside Mark and update after each Mark release, so a skill never asks for something the live Mark lacks.

Which way to connect

You work in	Use	Why
One Mark workspace	The plugin	One install gives you the connection and every skill.
Two or more Mark workspaces	Skills plus one MCP connection per workspace	The plugin carries one connection. Installing skills on their own lets you add a connection for each workspace.
A client without plugins (ChatGPT, Windsurf, Grok)	MCP connection, plus skills if your agent reads skill folders	Skills install into any agent that reads skill folders.

If you run Mark for several clients, as an agency or a consultant, use the second row.

The plugin

The plugin works in Claude Code, Cursor, Grok Bot, Codex, and VS Code (through Agent Plugins).

Claude Code

claude plugin marketplace add Life-With-Data/mark-agent-plugin

claude plugin install mark@mark-gtm

Then run /mcp, select mark, and choose Authenticate. Sign in to Mark and pick the workspace. Skills show up as /mark:plan-week and so on.

Cursor

Mark is not in the Cursor Marketplace yet. Until it is, install the plugin from the repo into Cursor's local plugin folder:

git clone https://github.com/Life-With-Data/mark-agent-plugin ~/.cursor/plugins/local/mark

Restart Cursor, or run Developer: Reload Window. Then go to Settings → MCP and choose Connect next to mark. Sign in to Mark and pick the workspace. The plugin also adds a rule that tells Cursor's agent when to use Mark. To update, run git pull in that folder.

One workspace per connection

The plugin connects to https://app.use-mark.com/mcp, and one connection works in one workspace. You cannot install the plugin a second time for another workspace. To change workspace, disconnect and authenticate again.

MCP and skills separately

Install the skills once, then add one MCP connection for each workspace. The same skills then work in every workspace you connect.

1. Install the skills

npx skills add Life-With-Data/mark-agent-plugin -g

-g installs them for your user, so they work in every folder. Leave it off to install into the current project only. The command asks which agents to install to (Claude Code, Cursor, Codex and others). To install one skill, add --skill plan-week. To update later, run npx skills update.

In Claude Code, skills installed this way run as /plan-week and so on, without the mark: prefix.

2. Add a connection for each workspace

Each workspace has its own MCP address that only works for that workspace:

https://app.use-mark.com/mcp/o/‹workspace id›

Copy it from the Mark app: open the workspace, go to Settings → AI agents, and copy This workspace only. An agent already connected to the workspace can also tell you: ask "What's this workspace's MCP address?"

Add one connection per workspace, named after the workspace. In Claude Code:

claude mcp add --transport http mark-acme https://app.use-mark.com/mcp/o/‹acme workspace id›

claude mcp add --transport http mark-globex https://app.use-mark.com/mcp/o/‹globex workspace id›

Then run /mcp and authenticate each one. On the Mark sign-in screen, pick the workspace that matches the address. If you pick a different one, the agent's calls fail with a message telling you to reconnect.

For other clients, follow Connect your AI agent and paste each workspace's own address in place of the shared one. In Claude and Claude Desktop, add one custom connector per workspace, named like "Mark – Acme".

Use the workspace-only address for every connection, not the shared /mcp address. Each one is locked to its workspace, so a connection can't drift to the wrong client's workspace.

3. Name the workspace when you ask

With several connections, say which workspace you mean: "Plan next week's LinkedIn for Acme." Each skill checks which connections exist, uses the one for the workspace you named, and asks if you didn't name one. It then confirms the workspace with Get Workspace before it does anything.

Skills

Skill	Job	Example prompt
get-started	See your workspace, what setup is left, and which skill to try first	"Help me get started with Mark."
plan-week	Draft the week's posts from pillars and the calendar	"Plan next week's LinkedIn from our pillars."
performance-review	Read search and content results, then propose (not save) changes	"What changed in search last month?"
ai-visibility	See how assistants see you and draft FAQ answers	"How do AI assistants talk about us?"
site-fixes	Read the audit and readiness. Mark never edits the site	"What should we fix on the site first?"
keyword-ideas	Research, shortlist, and save a brief as a draft	"Find keyword ideas for ADU payback."
draft	Draft a post or page for one channel with your brand's rules and Mark's checks, then save it for review	"Write a LinkedIn post about our new pricing." or `/mark:draft linkedin`
brand-interview	Set up your brand voice, phrases to avoid and claims from your own files and a short interview, saving each one only after you approve it	"Set up our brand rules in Mark from this repo."
review	Show posts waiting on you, say what blocks approval, then approve, request changes or reject each one only after you choose	"What's waiting on my review?"

Run brand-interview from your brand's own folder, so it can read the files that describe your brand (such as AGENTS.md, README.md and context/). It never reads .env files. For a client, run it from that client's folder and name the client's workspace.

When you state a rule while drafting, such as "we never say game-changer", draft asks before saving it as a rule for your whole workspace. Say no and it applies to that piece only.

Skills need a Mark MCP connection. Without one, a skill stops and links to Connect your AI agent.

On this page