Skip to content
Last updated

Getting Connected

Status: Closed Beta — You must be enrolled in the Infillion Agent Connector beta program to use this integration. Contact your Client Services representative for access.

Prerequisites

Before connecting to the Infillion Agent Connector, ensure you have:

  1. A MediaMath Platform account with active credentials
  2. Beta enrollment confirmed by your Client Services representative
  3. An MCP-compatible AI client — any client that supports the Model Context Protocol, including:
    • Anthropic Claude (claude.ai, Claude Desktop, or Claude Code)
    • Cursor IDE
    • VS Code with MCP extension
    • Custom AI agents using an MCP SDK
    • Other MCP-compatible clients

Connection Overview

The Infillion Agent Connector uses a Server-Sent Events (SSE) transport, which is the standard MCP connection method for remote servers. Your AI client connects to the MediaMath MCP server endpoint over HTTPS, authenticates using your credentials, and then maintains a persistent connection through which tool calls and responses flow.

Your AI Client ──── HTTPS/SSE ────► MediaMath MCP Server
                                         │
                                         ▼
                                    MediaMath Platform
                                    (Campaigns, Reporting,
                                     Creatives, etc.)

Connecting Your AI Client

Claude (claude.ai)

If you are using Claude through claude.ai or the Claude app, the Agent Connector can be added as a connected MCP server through your account's integration settings. Your Client Services representative will provide the specific server URL and any required configuration details during onboarding.

Claude Desktop / Claude Code

For Claude Desktop or Claude Code, add the MCP server to your configuration file:

{
  "mcpServers": {
    "mediamath": {
      "type": "url",
      "url": "<MCP_SERVER_URL_PROVIDED_DURING_ONBOARDING>"
    }
  }
}

Your Client Services team will provide the exact server URL during onboarding.

Cursor / VS Code

MCP server connections can be configured in your IDE's MCP settings. Add the MediaMath server using the SSE URL provided during onboarding. Refer to your IDE's MCP documentation for specific configuration steps.

Custom AI Agents

If you are building a custom AI agent or orchestration system, you can connect using any MCP SDK that supports the SSE transport. Popular options include:

  • TypeScript/JavaScript: @modelcontextprotocol/sdk
  • Python: mcp package

Your agent should connect to the provided SSE endpoint. The MCP server handles authentication and exposes the full set of available tools upon connection.

Authentication & Permissions

The Agent Connector authenticates through the MCP server using your MediaMath credentials. All operations performed by the AI agent are scoped to the same permissions as your MediaMath user account. This means:

  • The agent can only access organizations, agencies, and advertisers you have permission to view or edit.
  • Campaign creation, strategy updates, and other write operations require the appropriate role (e.g., Trader with edit access, Manager, or Admin).
  • Read-only users will be able to discover and inspect entities but not modify them.

No separate API key or OAuth2 client setup is required for the MCP connection — authentication is handled as part of the MCP server handshake.

Verifying Your Connection

Once connected, you can verify that the integration is working by asking your AI agent to perform a simple operation:

  • "Run a healthcheck on the MediaMath API" — This calls the healthcheck tool and confirms API connectivity, latency, and connection pool status.
  • "What is my MediaMath user info?" — This retrieves your user profile and confirms your identity and permissions.
  • "List the organizations I have access to" — This confirms your entity-level access scope.

If any of these operations fail, check your connection URL and ensure your beta enrollment is active. Contact Client Services for troubleshooting assistance.

Rate Limits & Usage

The MCP integration uses the same underlying MediaMath APIs and is subject to the same rate limits and usage policies. In practice, AI agents tend to make efficient, targeted API calls, so rate limits are rarely a concern for normal usage patterns.

If your agent workload involves high-volume batch operations (e.g., updating hundreds of strategies in a loop), coordinate with your Client Services team to ensure your usage profile is supported.

Security Considerations

The Agent Connector follows MediaMath's standard security practices:

  • All communication occurs over HTTPS/TLS.
  • Authentication tokens are managed server-side and never exposed to the AI client.
  • Operations are scoped to your user's existing permission set — the MCP connection cannot escalate privileges.
  • Audit trails for all operations are maintained in the standard MediaMath platform logs.

For more on MediaMath API security practices, see the security guidance in Best Practices.

Next Steps

  • Review the Tool Reference to understand what operations are available.
  • See Best Practices for recommended workflows and prompt patterns.
  • Contact your Client Services representative with any questions about your onboarding.
Previous page
Next page