Skip to main content

​
Overview

Criteo MCP is the connection layer between your AI application or agent environment and Criteo data and capabilities. This guide is for partners who want to connect an MCP-compatible AI application or agent environment to Criteo MCP and understand how to set it up. Typical users include:
  • Agencies, brands, retailers, and third-party platforms using MCP-compatible AI applications
  • Teams building proprietary agent environments that need access to Criteo capabilities
  • Technical owners responsible for setup and configuration
  • Teams with access to Enterprise LLM application licenses and business process owners defining workflows to be supported in their agentic platform

​
Why MCP?

Criteo MCP sits on top of Criteo APIs and exposes capabilities in a way that is easier for AI applications and agents to use. In practice, this helps partners by:
  • Reducing bespoke integration work between the AI application and individual APIs.
  • Making structured capabilities easier to discover and invoke.
  • Supporting natural-language workflows that combine retrieval, explanation, and next-step reasoning.
Partners may still use raw APIs directly where appropriate; MCP is seen as complementary to this existing capability.
MCP is most useful when the goal is to power AI-assisted workflows requiring LLM reasoning.

​
Using AI Tools with Criteo

When you connect your own AI assistant or agent to the Criteo platform (including via MCP), please keep the following in mind. AI-generated outputs — reporting summaries, recommendations, campaign settings, targeting or budget suggestions — may be inaccurate, incomplete, or out of date. Treat them as suggestions and review them before you act. Anything done through the connection, including creating or funding campaigns and the spend that follows, counts as your action and is your responsibility, whether a person or your AI agent initiated it. That includes keeping your credentials secure, deciding what permissions to grant, and making sure your campaigns and creatives comply with applicable law and Criteo’s policies. Criteo doesn’t provide or control your AI tool or its model provider, and gives no warranty for them. Connecting them means your Criteo data leaves our platform, so please make sure you have the rights to share it. Criteo gives no warranty as to AI outputs and is not liable for decisions made in reliance on them, subject to the limits in your agreement with us.

​
Get Started

​
Prerequisites

Before connecting to Criteo MCP, make sure you have:
  • A Criteo API application and valid credentials (provisioned by Criteo)
  • An MCP-compatible client, AI application, or agent platform that supports MCP integrations and can be configured
  • A shortlist of workflows you want to test first
  • Named technical and business owners on your side

​
Environments and Access

Criteo currently provides the following MCP environment:

Environment

MCP URL

Typical use

Production

https://mcp.criteo.com/mcp

Partner testing on approved live environments and approved production credentials

Criteo will share with you an endpoint collection (flavor) fitting your use case.
Ensure your agentic platform supports MCP integrations and that you are able to configure them. Before testing, confirm that the organization and credentials you plan to use have been approved for MCP access.
For complex authorization behavior, please reach out to your Criteo point of contact.

​
Supported Integration Patterns

Criteo MCP can be used through several integration patterns, depending on your needs. Please see more in Which Method Should You Choose? on the Access Flows page. Validated off-the-shelf AI application examples from current testing include: Partners may also connect their own agent environments or orchestration layers, provided they support MCP-compatible connection patterns.

​
Step-by-Step Integration Flow

1

Create or Obtain Credentials

Reach out to the Criteo team to validate and create credentials for your use case.
2

Confirm Access

Make sure the organization and credentials you plan to use are approved for the environment you want to test; this includes organization-level MCP access approval.
3

Configure the MCP Endpoint in Your Client

Add the relevant MCP server URL and authentication settings to your MCP-compatible client or agent environment.
4

Validate the Connection

Start with a simple scope or discovery test, such as:
  • Confirm which accounts are visible.
  • Confirm which tools are accessible.
  • Confirm that the returned scope matches the expected user, brand, or organization access.
5

Move into First Workflow Testing

Once setup is validated, begin testing a few agreed workflows rather than exploring every tool at once.

​
What Good Looks Like

A successful first integration should allow a partner to:
  • Connect to the MCP server without major setup blockers.
  • Retrieve the correct account, retailer, or brand scope.
  • Complete meaningful first workflows without manually stitching together raw API calls.
  • Receive outputs that are accurate, understandable, and useful to the intended user.
  • Clearly identify where the MCP is already valuable and where coverage is still missing.