MCP — Cursor

Connect Cursor to your Pad Cloud workspaces so the in-editor agent can read and update tasks, ideas, plans, and docs as you code.

Prerequisites

  • Cursor with MCP support (any current release)
  • A Pad Cloud account (or be willing to create one during the OAuth flow)

Steps

1. Open MCP settings

In Cursor, open the command palette (⌘⇧P / Ctrl-Shift-P) and run “MCP: Add Server”. Or navigate to Settings → Features → MCP servers and click Add server.

2. Choose “Remote (Streamable HTTP)”

Cursor supports both stdio (local subprocess) and remote (HTTP) MCP servers. Pick Remote.

3. Paste the Pad MCP URL

https://mcp.getpad.dev

Give the connection a name — Pad is what we recommend so log lines and Cursor’s UI both read clearly.

4. Sign in to Pad

Cursor opens an OAuth window. Sign in to your Pad account, or create one — the registration page on app.getpad.dev preserves your in-flight OAuth flow so you can sign up and continue the connection without restarting.

5. Pick workspaces and permission tier

The consent screen lists your Pad workspaces. Tick the ones Cursor can access; pick a tier (read-only / read-write / admin). Click Allow access.

Tier is capped by your role in each workspace — see the index page for the full model.

6. Verify

In a Cursor chat, ask:

“Set the Pad workspace to my-project and show me what’s on my plate.”

The agent should call pad_set_workspace followed by pad_project.action: next and report the recommended next task.

Pairing with the local /pad skill

Cursor reads project-local skills out of .agents/skills/. If you also pad agent install cursor from your project, the agent gets both the conversational /pad skill (works against any local config) AND the remote MCP connector (works against your Pad Cloud workspaces). They cooperate: the skill knows the conventions and patterns, the MCP server provides the typed tool surface.

What’s next

← All MCP setup guides · Troubleshooting →