Mark for AI agents

Connect Claude, ChatGPT, Grok, Cursor, VS Code, Codex, Windsurf and other AI agents to your Mark workspace. From your agent you can plan and draft posts, send them to review, approve what's waiting on you, check how content and search are performing, see how AI assistants find you, and research keywords. The agent acts as you, in the one workspace you pick when you sign in.

This repo is:

• the mark plugin: Mark's MCP server plus five skills, for Claude Code, Claude, Cursor, VS Code, Codex, Grok Bot and other clients that read Claude or Agent Plugins manifests;
• the mark-gtm marketplace that lists it.

The Mark MCP server (https://app.use-mark.com/mcp) is live. 63 tools. Sign in, pick a workspace, click Allow.

Full guide: Use Mark from AI agents

Skills

Every skill stops at explicit checkpoints before it creates, sends, spends credit or approves anything. In Claude Code they run as /mark:plan-week and so on, or trigger automatically from what you ask.

Tools

63 tools on the live server (tools.json), in these groups:

• Workspace and brand
• Strategy and brand
• Content: posts, review, calendar
• Assets
• Performance
• Keywords and competitors
• Websites and site health
• AI visibility
• Connections
• Connections and setup

Install

Server URL for every client: https://app.use-mark.com/mcp. On first use you sign in to Mark, choose a workspace, and click Allow. There are no API keys.

Claude Code: plugin (MCP + skills)

claude plugin marketplace add Life-With-Data/mark-agent-plugin
claude plugin install mark@mark-gtm

Then run /mcp, select mark, and choose Authenticate. MCP only: claude mcp add --transport http mark https://app.use-mark.com/mcp.

Claude (claude.ai, Claude Desktop)

Go to Settings → Connectors → Add custom connector, name it Mark, paste the URL, choose Connect, sign in, and pick your workspace. On Team and Enterprise plans an owner adds it under organization connectors. On plans with plugins, you can add this repo as a plugin marketplace instead to get the skills too.

ChatGPT

Go to Settings → Apps & Connectors → Advanced, turn on Developer mode, then choose Create. Enter the name Mark and the URL, and set authentication to OAuth. Sign in and pick your workspace. Add Mark to a chat from the + menu. On Business and Enterprise, an admin must allow custom connectors.

Grok (grok.com)

Go to grok.com/connectors → New Connector → Custom, paste the URL, and sign in. On Business and Enterprise, an admin provisions the connector in the xAI console first.

Grok Bot

Tell your bot: "Add a custom MCP server called mark at https://app.use-mark.com/mcp". Confirm, then click Authorize on the connect card and pick your workspace. Team admins may need to allow the URL in the MCP allowlist.

Cursor

Add to Cursor, or add this to ~/.cursor/mcp.json:

{ "mcpServers": { "mark": { "url": "https://app.use-mark.com/mcp" } } }

Then go to Settings → MCP and choose Connect next to mark. For the skills, see Skills only below.

VS Code (GitHub Copilot)

Install in VS Code, or add this to .vscode/mcp.json:

{ "servers": { "mark": { "type": "http", "url": "https://app.use-mark.com/mcp" } } }

Choose Start above the server entry, then sign in.

Codex (CLI, IDE extension, app)

codex mcp add mark --url https://app.use-mark.com/mcp
codex mcp login mark

Or add this to ~/.codex/config.toml:

[mcp_servers.mark]
url = "https://app.use-mark.com/mcp"

Windsurf

Add this to ~/.codeium/windsurf/mcp_config.json, then refresh the MCP panel and sign in:

{ "mcpServers": { "mark": { "serverUrl": "https://app.use-mark.com/mcp" } } }

Muse (Meta)

Muse doesn't document MCP support yet. Muse Code supports remote MCP servers: add an mcp_servers entry named mark with transport streamable_http and the URL above, then run muse mcp login mark. We'll update this section when the Muse app supports custom connectors officially.

Other MCP clients

Any client that supports remote MCP over streamable HTTP with OAuth works with the URL above.

Skills only (any agent)

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

The skills need the Mark MCP connection above to do anything.

Safety

• Your permissions, one workspace. The agent can do what you can do in Mark's UI, in the workspace you chose at sign-in, and nothing else. To switch workspaces, reconnect. If you leave the workspace or are removed from it, access stops within 30 seconds.
• Consequential tools are marked. Approving a post (approved posts publish at their scheduled time), deleting and paid lookups are annotated as destructive or open-world, so your client asks before running them. Approvals must name the exact revision the agent read. If the post changed since, Mark refuses.
• Everything is attributed. Actions show in Mark's review history as yours, "via ‹agent›".
• Skills add checkpoints. They ask before drafting, sending to review, spending credit or approving.

Repository layout

.claude-plugin/plugin.json       Claude plugin manifest (name: mark)
.claude-plugin/marketplace.json  Claude marketplace (name: mark-gtm)
.mcp.json                        Claude MCP config (remote HTTP)
plugin.json, mcp.json            Agent Plugins 1.0 manifests (same plugin, other clients)
skills/<name>/SKILL.md           The five skills (Agent Skills format)
tools.json                       Lockfile of 56 Mark MCP tool names, titles and annotations
scripts/check-skills.mjs         Checks skills, tool references and manifests (run in CI)

Development

node scripts/check-skills.mjs                       # skills + tool names + manifests
npx @anthropic-ai/claude-code plugin validate --strict .claude-plugin/marketplace.json
npx @anthropic-ai/claude-code plugin validate --strict .claude-plugin/plugin.json
npx @anthropic-ai/claude-code plugin validate --strict skills

When a tool is added or renamed in Mark, update tools.json in the same change, and CI fails if a skill references a tool that doesn't exist. Bump version in .claude-plugin/plugin.json, .claude-plugin/marketplace.json and plugin.json together.

Issues and security reports: see SECURITY.md.

License

MIT © Life With Data