Setting up your AI assistant¶

Prerequisites¶

A u-Slicer account. The MCP server wraps the u-Slicer API and reuses your own credentials, so it can only reach the data you already have access to.
An MCP-aware AI assistant, for example Visual Studio Code with GitHub Copilot, Claude Code, Claude Desktop or Cursor.
Access to an LLM through that assistant. The MCP server provides the tools and the data; the model interprets your question, calls the tools and writes the answer. The more capable the model, the better the result.

Server endpoint¶

Property	Value
MCP endpoint	`https://uslicer.criteo.com/mcp`
Transport	Streamable HTTP
Authentication	OAuth 2.0, browser sign-in with your u-Slicer account

Note

If your organization uses a different u-Slicer host, the endpoint is that host with /mcp appended. Everything else stays the same.

Visual Studio Code with GitHub Copilot¶

Create a .vscode/mcp.json file in your workspace:

{
  "servers": {
    "uslicer": {
      "type": "http",
      "url": "https://uslicer.criteo.com/mcp"
    }
  }
}

Reload the window, open Copilot Chat in Agent mode and the uslicer tools appear in the tool picker. See the VS Code MCP documentation for details.

Claude Code (CLI)¶

Run in a terminal:

claude mcp add --transport http uslicer https://uslicer.criteo.com/mcp

Then check the connection inside Claude Code with /mcp.

Claude Desktop¶

Add the server under Settings → Connectors → Add custom connector, using the endpoint https://uslicer.criteo.com/mcp.

Cursor¶

Remote HTTP servers are not connected reliably out of the box; bridge the server through mcp-remote in your MCP configuration:

{
  "mcpServers": {
    "uslicer": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://uslicer.criteo.com/mcp"]
    }
  }
}

This requires Node.js to be installed.

Other clients¶

Any client that speaks MCP over Streamable HTTP should work. What it needs to know:

Server URL: https://uslicer.criteo.com/mcp
Transport: Streamable HTTP
Authentication is discovered through the OAuth Protected Resource metadata endpoint (RFC 9728): https://uslicer.criteo.com/.well-known/oauth-protected-resource/mcp
Advertised scopes: openid, email, profile, offline_access, uslicer

Signing in¶

The first time your assistant calls a u-Slicer tool, it starts an OAuth flow: a browser window opens, you sign in with your u-Slicer account and approve access. The client stores the resulting token and refreshes it on its own. When the token finally expires, the browser sign-in is repeated.

The server validates that token and forwards it to the u-Slicer API, so every request is executed as you: the same projects, slicers, columns and row-level restrictions as in the UI.

Verifying the setup¶

Ask your assistant:

Using u-Slicer, list the projects I have access to.

The assistant should ask for permission to run a u-Slicer tool (grant it) and then print your projects. Mentioning u-Slicer explicitly helps the model realize it should use the MCP tools rather than answer from memory.

Troubleshooting¶

Symptom	What to check
Server shows as disconnected	The endpoint URL, and that your network can reach `https://uslicer.criteo.com`.
Sign-in window never opens	Some clients only start the OAuth flow on the first tool call — ask a question that needs u-Slicer data.
`list_projects` returns nothing	Your account has no project access, or you signed in with the wrong account. Sign out in the client and reconnect.
Tools are not used at all	The assistant may be in a non-agent mode, or the tools are disabled in its tool picker. Say "use u-Slicer" in the prompt.
A project or column is missing	Permissions are inherited from your u-Slicer account. If you cannot see it in the UI, MCP cannot see it either.