One URL connects any MCP-speaking agent to the buckets you choose. Sign in, tick the buckets, done — nothing to install.
Every agent takes the same server address — give it this URL wherever your tool adds MCP servers:
https://mcp.copybucket.com/mcp
claude mcp add --transport http copybucket https://mcp.copybucket.com/mcp
In Claude, open Settings ▸ Connectors ▸ Add custom connector and paste the URL.
Add to Cursor with one click — or paste the URL into an entry in ~/.cursor/mcp.json.
Install in VS Code — agent mode picks it up directly.
{
"mcpServers": {
"copybucket": { "type": "http", "url": "https://mcp.copybucket.com/mcp" }
}
}
On first use, your AI sends you to copybucket.com to sign in and pick the buckets it may reach. Nothing is ticked for you: an agent sees exactly what you approve — and you can change or revoke that whenever you like.
Name the bucket in plain language — the tools (list, get, search, put, edit, move, copy, create, delete) do the rest.
Every agent you've approved is listed on your agents with access page and in the Mac app under Settings ▸ AI Agents — same list, same Revoke button. Revoking cuts the agent off immediately, mid-session included.
CopyBucket for Mac can also register a local MCP server: open Settings ▸ AI Agents and the guided flow sets it up with one click for Claude Code, Claude Desktop, Cursor, OpenCode, and Codex/ChatGPT Desktop. Agents on that Mac get direct file paths for large items, and it works offline for cached content.
You only need one of the two — the hosted server is the recommendation, and it never shows macOS permission prompts.
Visibility is off by default, bucket by bucket. Every item an agent adds is labeled with the agent's name, and the app shows you live when an agent is reading. Read the full story in the privacy promise.