How to Connect Clipzing to Claude, ChatGPT, and Cursor (MCP Setup in 2 Minutes)

Connecting Clipzing to an AI assistant takes about two minutes and there is no API key to copy anywhere. Browser assistants take a URL; desktop apps and editors take a one-line config. This post has the exact steps for all nine supported apps, including the restart quirks that account for most "it isn't working" reports.

Before you start

You need a Clipzing subscription. Agent access starts at the Starter plan — the unsubscribed free tier has no MCP access at all, deliberately, because an API key outlives a session and issuing one to a non-customer is the worst version of a leaked credential.

For the local clients you also need uvx, which ships with uv. If you don't have it:

curl -LsSf https://astral.sh/uv/install.sh | sh

Browser assistants need nothing installed.

Claude on the web

1. Open Settings → Connectors → Add custom connector.
2. Enter the URL: https://clipzing.com/api/v1/mcp/
3. Claude redirects you to Clipzing. Sign in and pick which permissions to grant.
4. Approve. Connected.

That's the whole flow — the permission screen is where you choose scopes, and starting with read-only is a reasonable first move.

Claude Desktop

1. Settings → Developer → Edit Config. This opens your config file:
   • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
   • Windows: %APPDATA%\Claude\claude_desktop_config.json
   • Linux: ~/.config/Claude/claude_desktop_config.json
2. Add Clipzing to the mcpServers block:

{
  "mcpServers": {
    "clipzing": {
      "command": "uvx",
      "args": ["clipzing-mcp"]
    }
  }
}

3. Quit Claude completely and reopen it.

On first use, the server opens a pairing page in your browser. Click approve while signed in to Clipzing and the credential writes itself.

Claude Code

One command:

claude mcp add clipzing -- uvx clipzing-mcp

Then start a new session — running sessions keep the tool list they started with.

Claude Code is also the easiest place to install the Clipzing skills, which live in ~/.claude/skills/. Each skill is its own folder with a SKILL.md inside — for example ~/.claude/skills/clip-critique/SKILL.md. For a single project, use .claude/skills/ in the project root instead.

ChatGPT on the web

Developer mode is a genuine prerequisite here, not a nicety — without it there's nowhere in the UI to add a server at all.

1. Settings → Security and login → turn on Developer mode.
2. Go to Plugins → + and add https://clipzing.com/api/v1/mcp/
3. When asked how to authenticate, choose OAuth.
4. Sign in to Clipzing and pick permissions.

ChatGPT desktop app

The desktop app shares Codex's config file, and that file is TOML, not JSON. Handing it a JSON block produces a config that fails silently.

1. Settings → MCP servers → Add server → choose STDIO → Save → Restart.
2. Add to ~/.codex/config.toml (Windows: %USERPROFILE%\.codex\config.toml):

[mcp_servers.clipzing]
command = "uvx"
args = ["clipzing-mcp"]

3. Use the explicit Restart button in the MCP settings panel.

Cursor, Windsurf, Zed, and Cline

All four take the same local server, in each one's own config location and key name — Zed, notably, uses context_servers rather than mcpServers, which trips people who copy a Cursor config across.

Rather than reproduce four slightly-different blocks here and let them drift, the setup guide has the exact path and config for each, verified per client. Pick your app and copy the block it shows you.

Verify it worked

Ask your assistant:

"List my Clipzing clips."

You should get a table of real clips with titles, durations, and virality scores. If you do, everything downstream works.

If you get nothing, work down this list:

Tools not showing at all — restart properly. Full quit for Claude Desktop, new session for Claude Code, the Restart button for ChatGPT desktop. This is the cause the large majority of the time.

"No agent access" or similar — check your plan. Starter and above only.

Tools show but writes fail — check scopes. A read-only key can list and analyze but cannot change anything, which is the intended behavior, not a bug.

Publishing "doesn't work" — it's waiting on you. Publish, schedule, delete, and batch operations return a summary and stop for confirmation. Read the summary and tell the assistant to proceed.

Choosing scopes

Four scopes, and it's worth being deliberate rather than granting everything on day one.

A read-only key for the first week costs you nothing and tells you a great deal. The assistant can still critique every clip you make, spot framing problems with watch_clip, and draft titles — you just apply them yourself.

Revoking access

Delete the key from your Clipzing account. The connection dies immediately and the assistant simply stops seeing the tools. There's no cleanup on the client side beyond removing the config entry if you want it tidy.

Frequently asked questions

Do I need to be technical? For browser assistants, no — it's a URL and an approval click. For desktop apps, you paste one block into one file.

Can I connect several apps at once? Yes. Each gets its own credential. Starter allows one key; Creator allows five.

Does this work with Gemini? Not yet — Gemini's assistant apps don't support remote MCP connectors the way Claude and ChatGPT do. When they do, this works with no change on our side.

Where's the server URL again? https://clipzing.com/api/v1/mcp/