Connect Your AI With MCP
Snipzr runs a remote MCP server at https://mcp.snipzr.com/mcp. Once your client is connected,
your assistant can create links on your domain, list and update them, read cookie-less stats and
campaign reports, check your quota, build UTM links and render QR codes, asking you first before
anything destructive. There is nothing to install: it is one hosted endpoint. Clients that sign in
with Snipzr, such as Claude, ChatGPT, Claude Code and Codex, work on every plan, Free included, where
the Free allowance of 30 new links a month applies. Tokens, for clients that take a header instead,
need a paid plan.
Get a token
- Open Settings and find User API tokens.
- Select Create token and name it after the client, for example
Claude Code. Pick the workspaces it may act in and untick any scope the client should not have there. A read-only assistant needs nothing beyond the read scopes. - Copy the secret. It is shown exactly once; if you lose it, roll the token for a fresh one.
A workspace token (from the Workspace API tokens card) works too: it acts as that one workspace and is the right choice for a shared team assistant.
One token per client is the healthy habit: when a laptop or a client goes away, you revoke just that token and everything else keeps working.
Connect your client
Claude Code, two commands; the second opens the Snipzr sign-in:
claude mcp add --transport http snipzr https://mcp.snipzr.com/mcp
claude mcp login snipzr
Prefer a token? Add --header "Authorization: Bearer YOUR_TOKEN" to the first command and skip the
second.
Claude on the web, desktop and mobile connects from Claude's directory, no token needed:
- Open Snipzr in Claude's directory and select Connect to Claude.
- Claude opens a Snipzr sign-in page where you approve exactly what the connector may do. Every permission is a checkbox you control, and you can revoke the whole connection any time under Settings, then User API tokens.
- Prefer to add it by address? In Claude, open Settings, then Connectors, choose Add,
then Add custom connector, and enter
https://mcp.snipzr.com/mcp. - Prefer a pasted token? Add it by address and use the Request headers option (beta): set
Authentication to None and add the header
authorizationwith the valueBearer YOUR_TOKEN, including the wordBearerand the space.
ChatGPT on the web connects in developer mode, which OpenAI offers on Plus, Pro, Business, Enterprise and Education accounts:
- Open Settings, then Security and login, and turn on Developer mode.
- Open Plugins, select the plus button and create a developer-mode app with the server address
https://mcp.snipzr.com/mcpand OAuth as the authentication. ChatGPT opens the Snipzr sign-in where you approve exactly what it may do. - In a conversation, choose Developer mode from the plus menu and select Snipzr. ChatGPT asks before every write and never before a read.
Codex CLI, the Codex IDE extension and the ChatGPT desktop app share one configuration, so two commands set up all three; the second opens the Snipzr sign-in:
codex mcp add snipzr --url https://mcp.snipzr.com/mcp
codex mcp login snipzr
In the desktop app the same entry lives under Settings, then MCP servers, then Add
server (Streamable HTTP, the address above), followed by Authenticate. Prefer a token? Set
bearer_token_env_var on the server entry in config.toml to the name of an environment variable
that holds it.
Cursor, VS Code and any other MCP client take the generic remote config:
{
"mcpServers": {
"snipzr": {
"url": "https://mcp.snipzr.com/mcp",
"headers": { "Authorization": "Bearer YOUR_TOKEN" }
}
}
}
In Cursor this lives in .cursor/mcp.json; in VS Code, in your mcp.json, where GitHub
Copilot's agent mode picks it up. A client that only speaks stdio can bridge with
npx mcp-remote https://mcp.snipzr.com/mcp.
To confirm it worked, ask your assistant to list the Snipzr tools. Thirteen names starting with
snipzr_ means you are connected.
The thirteen tools
| Tool | What it does | Scope |
|---|---|---|
snipzr_create_link | Create one short link: destination, optional domain, back-half, title | links:write |
snipzr_create_links_batch | Create up to 100 links in one call | links:write |
snipzr_list_links | List links with text, domain and campaign filters | links:read |
snipzr_get_link | Read one link by its back-half | links:read |
snipzr_update_link | Change a destination or title | links:write |
snipzr_delete_link | Delete a link, with a confirmation gate | links:write |
snipzr_get_link_stats | Per-day clicks and scans, with bots, AI assistants and suspected automated traffic counted separately | stats:read |
snipzr_get_campaign_report | Clicks rolled up by campaign | stats:read |
snipzr_get_usage | Plan, remaining allowance and reset date | usage:read |
snipzr_list_domains | Your short domains and their status | domains:read |
snipzr_get_qr_code | A scannable PNG QR code for an existing link | links:read |
snipzr_build_utm_link | Compose a UTM link that follows your conventions | links:read |
snipzr_list_workspaces | The workspaces the connection can act in, with the id to pass to other tools | any |
Every tool carries read-only or destructive annotations, so clients ask before a delete and never before a read. Links an agent creates count toward the same monthly allowance as links you create by hand, and the server tells the agent plainly when the allowance runs out.
Good to know
- The server acts only within the scopes on the token; everything else is refused with a message that names the missing scope.
- A user token acts in your personal workspace by default. To work in another workspace on the
token, the assistant passes that workspace's id as the
workspaceIdargument of any tool; the server checks your live membership there on every call, so leaving a workspace cuts the token's access to it instantly. - Revoking a token in Settings cuts the client off immediately.
- The server never logs your destinations, and analytics stay cookie-less.
- The full setup page with per-client details lives at snipzr.com/integrations/mcp.