Xandr MCP – Getting started

The Xandr MCP provides read-only access to Xandr data through the Model Context Protocol (MCP). It is designed for AI clients that need to discover, retrieve, and summarize data from Xandr without making platform changes.

The service currently supports the following Xandr product surfaces:

Monetize: https://monetize.xandr.com/mcp
Curate: https://curate.xandr.com/mcp

These endpoint URLs are stable and are not versioned. They will not change.

Important

Xandr MCP is currently in closed beta. Tool coverage, schemas, and behavior may change without notice.

Introduction

Xandr MCP is intended for data retrieval, discovery, and analysis workflows. MCP-compatible clients can:

Query reporting and analytics data
Inspect object metadata and platform entities
Retrieve reference and lookup data
Review change history and saved report information
Access selected planning, forecasting, or troubleshooting experiences (where enabled)

Xandr MCP is currently a read-only product. It does not create, update, or delete Xandr objects.

Write operations may be added in the future but would require explicit enablement.

Using Xandr MCP

To connect an MCP client:

Configure a remote HTTP MCP server that points to either the Monetize or Curate endpoint
Provide a valid Xandr API authentication token

Authentication

Xandr MCP requires a valid Xandr API authentication token. Token format, lifetime, refresh behavior, and all other authentication mechanics are defined by the Xandr Authentication Service.

MCP does not introduce a separate authentication model—it uses the same token flow as the standard Xandr API.

Authentication Service

To authenticate requests, include the token in the Authorization header:

Authorization: Bearer <your-xandr-token>

If the token expires, refresh it using the standard Xandr authentication flow and update it in your MCP client as needed.

Permissions and access control

Xandr MCP does not introduce a separate permission model.

Does not grant new platform access
Respects existing permissions assigned to the authenticated user or token
Enforces access through the underlying Xandr APIs

In practice, MCP can only expose data and capabilities that the authenticated identity is already authorized to use in Monetize or Curate.

MCP does not enable platform functionality that is not already available to your account, member, or seat. For example, if forecasting is not enabled for your seat, forecasting-related MCP tools will not be available.

Configuration files

The configuration method depends on the client. Configuration files (such as mcp.json) define MCP servers, endpoints, and authentication settings. See the client-specific sections below for details.

VS Code uses mcp.json
Claude Code uses .mcp.json
Claude.ai and Claude Desktop use connector configuration in the UI instead of a file

mcp.json is a client-specific configuration file used to define MCP servers, endpoints, and authentication settings.

Other MCP-compatible tools may support different configuration models, authentication requirements, or setup flows. Refer to the documentation for your client.

VS Code

You can configure MCP servers in:

A workspace file at .vscode/mcp.json
Your user profile via the MCP: Open User Configuration command

Example `.vscode/mcp.json`

{
  "servers": {
    "xandr-monetize": {
      "type": "http",
      "url": "https://monetize.xandr.com/mcp",
      "headers": {
        "authorization": "Bearer ${input:xandrAuthToken}"
      }
    },
    "xandr-curate": {
      "type": "http",
      "url": "https://curate.xandr.com/mcp",
      "headers": {
        "authorization": "Bearer ${input:xandrAuthToken}"
      }
    }
  },
  "inputs": [
    {
      "id": "xandrAuthToken",
      "type": "promptString",
      "description": "Xandr API authentication token",
      "password": true
    }
  ]
}

VS Code also supports adding a server through the Command Palette (MCP: Add Server). However, a checked-in .vscode/mcp.json file is typically the clearest option for shared team setups.

After configuration:

Open the workspace in VS Code
Open .vscode/mcp.json or run MCP: Open User Configuration
Save the configuration
Allow VS Code to start the server when prompted
Use Chat—MCP tools are discovered automatically

If your client supports input prompts, this approach avoids hard-coding the authentication token in the file.

Claude Code

Claude Code supports project-scoped MCP configuration using a .mcp.json file at the project root. Its JSON structure differs from VS Code and uses mcpServers instead of servers.

Example `.mcp.json`

{
  "mcpServers": {
    "xandr-monetize": {
      "type": "http",
      "url": "https://monetize.xandr.com/mcp",
      "headers": {
        "Authorization": "Bearer ${XANDR_AUTH_TOKEN}"
      }
    },
    "xandr-curate": {
      "type": "http",
      "url": "https://curate.xandr.com/mcp",
      "headers": {
        "Authorization": "Bearer ${XANDR_AUTH_TOKEN}"
      }
    }
  }
}

Claude Code also supports adding MCP servers using the CLI instead of manually editing the JSON file.

Example CLI commands

claude mcp add --transport http --scope project xandr-monetize https://monetize.xandr.com/mcp --header "Authorization: Bearer your-token"
claude mcp add --transport http --scope project xandr-curate https://curate.xandr.com/mcp --header "Authorization: Bearer your-token"

After adding the server:

Use /mcp inside Claude Code to inspect server status
Complete any required authentication flow prompted by the client

Claude.ai and Claude Desktop

For remote MCP usage in Claude.ai and Claude Desktop, configuration is done through a custom connector flow rather than a local mcp.json file.

Set up a remote connector

Open Customize > Connectors in Claude
Select Add a custom connector
Enter the MCP endpoint URL (for example, https://monetize.xandr.com/mcp or https://curate.xandr.com/mcp)
Complete authentication if prompted
Enable the connector for the conversation where you want to use MCP

For Team and Enterprise plans, connector creation may be restricted to organization owners or admins.

Note

Claude Desktop also supports local MCP servers through claude_desktop_config.json. However, current Anthropic guidance for remote MCP connectors uses the connector UI rather than mcp.json.

For a repository example with additional local and staging configurations, see mcp.json example above.

Rate limits

Xandr MCP requests are subject to rate limiting. The same throttling principles that apply to the Xandr API also apply to MCP.

For more information, see:

Throttling, Pagination, and Filtering

MCP usage consumes a portion of the API rate limits available to the authenticated client.

Typical interactive usage is unlikely to reach these limits
Automated or high-frequency integrations should handle throttling appropriately

Available tools

Xandr MCP exposes tools across the following functional areas:

Reporting and analytics: Run supported reports, retrieve dimensions and metrics, and analyze performance data
Report management: Check report status, download completed reports, and inspect saved report definitions
Object discovery and lookup: Search supported Xandr entities and inspect available object fields
Reference data: Retrieve lookup datasets such as geography, device, media, and other classification data
Change history: Review audit and change-log information for supported entities
Planning and forecasting: Access selected planning or forecasting data where available for your account
Troubleshooting: Use diagnostic tools for supported workflows where available

Typical reporting workflow

For reporting scenarios, a common workflow is:

Identify the report to use
Inspect available dimensions, metrics, and filters
Run the report with the required grouping, filters, and date range

Examples

Analyze Monetize inventory performance for the last 7 days by publisher or placement
Retrieve Curate deal metrics for a specific buyer or seller deal
Check provisional network analytics for same-day delivery trends
Inspect a report catalog to understand available columns and filters before running a query

Depending on the report, MCP may expose:

Direct report tools for commonly used report types
Discovery-style tools for exploring report metadata and executing queries

Typical object discovery workflow

For object exploration, a common workflow is:

Discover supported object types using xandr_list_objects
Inspect the schema and available filters for a specific object using xandr_describe_object
Query objects using xandr_search_object or an object-specific search tool

Examples

Discover whether advertiser, line-item, publisher, or deal is available as a supported object type
Inspect available fields and filters for a profile or publisher object before querying
Search for a specific advertiser, publisher, deal, or profile after identifying the appropriate filters

Some commonly used entities (such as advertisers or publishers) may have dedicated search tools for simpler and faster queries. Other entities use a general workflow where you first discover available object types, inspect their schema, and then query them using standard search tools.

Other common tool patterns

Reference data: Follows a discover → describe → search pattern for datasets such as geography, device, language, and media classifications
Report management: Check asynchronous report status, inspect saved reports, and download completed output
Change history: Review audit and change-log information for supported entities
Planning, forecasting, and troubleshooting: Available only when the corresponding platform capability is enabled for the authenticated user

Note

Tool names, schemas, coverage, and availability may change without notice during closed beta.

MCP and model responsibilities

When troubleshooting, distinguish between the responsibilities of the MCP server and the model or client.

MCP server responsibilities

Exposes tools
Validates inputs
Calls underlying Xandr APIs
Returns structured results or errors

Model or client responsibilities

Interprets the prompt
Selects which tools to use
Chooses parameters
Sequences tool calls
Summarizes results

Understanding different outcomes

Issues can originate from different layers:

Model or client layer: Incorrect tool selection, invalid filters, or misinterpretation of the prompt
MCP layer: Unexpected input validation failures, incorrect API calls, or inconsistent tool output
Platform layer: Access denied, missing data, or unavailable features based on account permissions

Similar prompts may produce different results across models or clients, even when using the same MCP server. This is because tool selection and reasoning are handled by the model, not MCP.

Troubleshooting guidance

For effective troubleshooting, capture:

What the user asked the model to do
What the model attempted through MCP

This helps determine whether the issue originates from prompt interpretation, model reasoning, client integration, MCP behavior, or the underlying platform capability.

Product scope

The current product scope is intentionally limited:

Read-only access: Supports data retrieval only
No write operations: Does not create, update, or delete objects
Future extensibility: Write operations may be added with explicit enablement
Closed beta: Availability is limited during rollout
Permission-bound access: Limited to existing authenticated permissions

Some capabilities may be available only for specific products, users, environments, or beta cohorts.

Support requests

When reporting an issue or requesting support, provide concise but specific information to help speed up investigation.

Include the following details:

Platform: Monetize or Curate
MCP client: For example, VS Code, Claude Code, or another MCP-compatible client
Model: If known
Original prompt: Or a close equivalent that reproduces the issue
Tool usage: Which tools the model attempted to use
Tool results: Whether each tool call succeeded, failed, or returned incomplete data
Relevant identifiers: Such as report type, object type, object ID, deal ID, advertiser ID, or date range
Errors or behavior: Exact error message or a summary of incorrect behavior
Reproducibility: Whether the issue is reproducible or intermittent

Keep the report concise, but include enough detail for others to understand and reproduce the issue without reconstructing the entire session.

Suggested support template

You can provide the following template to your LLM and ask it to summarize the session before submitting a support request.

# Xandr MCP support summary

## Issue summary
- Short description:
- Impact:
- Reproducible: Yes / No / Unknown

## Environment
- Platform: Monetize / Curate
- MCP client:
- Model:
- Environment: Production / Staging / Unknown
- Approximate time of issue:

## User intent
- Original prompt:
- Expected outcome:
- Actual outcome:

## Tool usage
| Tool | Purpose | Result | Notes |
|------|---------|--------|-------|
| tool_name | reason for use | Success / Failed / Partial | short detail |

## Relevant inputs
- Report type:
- Object type:
- IDs or names used:
- Date range:
- Filters or grouping:

## Errors or suspicious output
- Exact error message(s):
- If no formal error, describe the incorrect or incomplete result:

## Reasoning context
- Why the model chose those tools:
- Whether the model misunderstood the prompt, platform, or available capability:

## Additional context
- Whether the capability works outside MCP:
- Whether the account or seat has access to the underlying platform feature:
- Any additional information that may help reproduce or debug the issue:

Suggested instruction for your LLM

You can provide the following instruction to your LLM to generate a support summary:

Summarize this Xandr MCP issue for a support ticket. Be concise, but include:
- Platform  
- MCP client  
- Model (if known)  
- Original prompt  
- Tools used  
- Whether each tool succeeded or failed  
- Relevant IDs and date ranges  
- Exact errors  
- Any context needed to reproduce or debug the issue

Additional resources

Authentication Service

Last updated on 2026-04-16

Xandr MCP – Getting started

Introduction

Using Xandr MCP

Authentication

Permissions and access control

Configuration files

VS Code

Example .vscode/mcp.json

Claude Code

Example .mcp.json

Example CLI commands

Claude.ai and Claude Desktop

Set up a remote connector

Rate limits

Available tools

Typical reporting workflow

Examples

Typical object discovery workflow

Examples

Other common tool patterns

MCP and model responsibilities

MCP server responsibilities

Model or client responsibilities

Understanding different outcomes

Troubleshooting guidance

Product scope

Support requests

Suggested support template

Suggested instruction for your LLM

Additional resources

Additional resources

Example `.vscode/mcp.json`

Example `.mcp.json`