Interact with Agents, Mosaic Models, and Cubes Using an MCP Server

Beginning in Strategy (July 2026), universal agents are included in the response from the list_agent tool. ChatGPT and Copilot can use dynamic redirect URLs.

Beginning in Strategy One (March 2026), you can access Mosaic models in the MCP server out of the box, replacing the Strategy One (February 2026) preview feature. This allows you to ask questions in MCP directly against the unified semantic layer. For details, see Access Mosaic Models in the MCP Server.

Preview features are early versions of features and are not to be used in a production environment as the core behavior remain subject to change between preview and GA. By selecting to expose preview features, you can access and use them as you would any other functionality. The official versions of preview features are included in subsequent releases.

Beginning in Strategy One (December 2025), you can interact with agents from multiple clients in a ChatGPT-like interface. The MCP server in Strategy Library exposes tools that allows MCP client applications to discover, interact with, and query agents directly. MCP clients can list available agents, view their details, and direct questions to specific agents via remote calls. The MCP server enables real-time, cross-system interaction with agents from multiple clients, improving usability, extensibility, and integration across platforms. It reduces operational friction and increases automation by allowing external tools to leverage agents programmatically.

Overview

Interact with Strategy agents and query Mosaic Models and cubes from MCP clients such as Claude, Copilot, Gemini, ChatGPT, and more.

Strategy Library provides two Model Context Protocol (MCP) Server endpoints for AI agents that support the MCP Client Capabilities. Both endpoints use the same authentication and connection steps. They differ only in the endpoint URL and the tools they expose. You can add both connectors; they operate independently.

  • Agents MCP: Ask questions of the certified agents your organization has already built. Your MCP client discovers the available agents, routes the question, and the agent answers using its own data, business context, and instructions. See Agents MCP.

    • Link URL: https://env-domain.trial.cloud.strategy.com/collaboration/mcp/agent

  • Mosaic MCP: Query Mosaic models and AI-enabled cubes directly, with no agent required. Your MCP client reads the business context built into each mode and uses it to write accurate SQL. The context includes objects such as attributes, metrics, definitions, and descriptions. Strategy executes the query and returns governed results. See Mosaic MCP.

    • Link URL: https://env-domain.trial.cloud.strategy.com/collaboration/mcp/mosaic

In the URLs, replace domain with your MCP server.

For clients that use manual OAuth configuration, use the following values:

Connection URL
Authorization URL https://env-domain.trial.cloud.strategy.com/collaboration/authorize
Token URL https://env-domain.trial.cloud.strategy.com/collaboration/token
Refresh URL https://env-domain.trial.cloud.strategy.com/collaboration/token
Scopes offline_access

Getting Started with MCP

Connect to Strategy MCP by adding the server URL to your client and authenticating through Strategy.

  1. Enable the MCP Server. An administrator enables the MCP server once per environment and grants the required privileges.

  2. Pick your server. Use the agents MCP to ask questions through certified agents. Use the Mosaic MCP to query models and cubes directly. You can also connect both.

  3. Add the server URL to your client. Copy the URL for your server and add it to your client. Sign into Strategy Library when prompted. See Connect a Client Using an MCP Server for detailed steps for each client type.

  4. Ask your first question. Start a new chat and ask "What Strategy agents do I have?" (if you're using the Agents MCP) or "What data can I query?" (in Mosaic MCP). If the client answers with your agents or models, the connection works. See Using MCP Server.

User Privileges

  • Users must have the Use MCP Server privilege.

  • To use Agents MCP, users must have the Run AI Bots privilege.

  • Users cannot be administrators or members of the Platform Support Administrator user group.

Content Requirements

  • Agents must be active, certified, and added to the user's Library. Users must have view permissions on the agents. Certify only the agents you want to expose.

  • Mosaic models must be published and certified to be accessible through MCP. Users must have view permissions on the models. The Mosaic endpoint also provides access to live and in-memory intelligent cubes and in-memory MTDI cubes. These cubes must be enabled for AI and certified to be accessible through MCP.

Prerequisites

Most connection failures occur when these prerequisites are not met. The environment must:

  • Be publicly accessible

  • Have collaboration services enabled

If you still have issues, see Troubleshooting the MCP Server Integration for tips and solutions.

Using MCP Server

Both servers work the same way: your MCP client calls the tools below automatically as you chat. Knowing what each tool does helps you understand what each server can answer.

Agents MCP

The Agents MCP server exposes the following tools. The table below describes what each tool does.

Tool What It Does
list_agents Lists the agents you have access to, with each agent's name, description, and project. Universal agent descriptions also include their sub-agents. The client calls this first to pick the right agent for your question.
ask_agent Sends a question to a selected agent and returns a natural-language answer from its data. The client includes recent conversation history so follow-up questions work. If the question calls for a chart or table, the response also includes the data to render it.

You do not need to name an agent in your question. The client matches your question to the best agagentent using the descriptions from list_agents. You can also address an agent directly, for example, "Ask the Retail Agent about last week's returns."

Example Questions

  • List my Strategy agents and what each one covers.

  • What were the top 5 products by revenue last quarter?

Mosaic MCP

The Mosaic MCP server exposes tools that the client chains together to answer a question, moving from project to model to semantics to SQL.

Tool What It Does
get_projects Lists the projects you have access to. Usually the client's first call, to find where your data lives.
get_models Lists the certified models and datasets in a project, with names and descriptions. Each result is labeled as a Mosaic Model or an Other Model (MTDI and intelligent cubes).
get_semantics Returns a model's full definition: every attribute and metric, with data types, business descriptions, and metric formulas. The client reads this before querying so its SQL uses real column names and pre-defined metrics instead of guessing.
query Runs a read-only SQL query (Trino dialect) against the project's models and returns the results. Write and DDL statements are rejected.

Answers stay governed. Queries use the attributes and metrics defined in the model, so results match the definitions that your data team has already validated. Access is read-only; your data cannot be changed through MCP.

Example Questions

  • What data can I query?

  • Show revenue by region for the last 12 months.

Access Mosaic Models in the MCP Server

Beginning in Strategy One (June 2026):

  • Certify the Mosaic model to make it accessible through MCP. Requiring certification allows you to limit the number of models and cubes returned through MCP.

  • Access other AI-enabled cubes including live or in-memory intelligent cubes, and in-memory MTDI cubes through the Mosaic MCP endpoint.

Beginning in Strategy One (March 2026), this feature is available out of the box, replacing the preview feature. Ensure that the Use MCP Server privilege is enabled.

Beginning in Strategy One (February 2026), you can access Mosaic models in the MCP server as a preview feature.

This allows you to ask questions in MCP directly against the unified semantic layer, without needing to go through an agent. It also results in better performance because the interaction time with AI is saved.

You can use the following MCP clients to access Mosaic models:

When you connect to the MCP Server, use the following syntax for the MCP server URL endpoint:

https://env-domain/collaboration/mcp/mosaic

Replace domain with your MCP server.

You can use the following workflows with Mosaic models:

  • Get projects (get_projects) to list available projects

  • Get Mosaic models and other AI-enabled cubes (get_mosaic_models); you can filter by project

  • Get Mosaic models (only_mosaic_models) which returns Mosaic models only, no other cubes; you can filter by project

  • Get semantics for a table (get_semantics) which returns the full model metadata, including attributes, metrics, data types, descriptions, and underlying table/column information

  • Query to ask normal Agent questions such as create a graph with the data or provide a summary (ask_agent) and receive the agent's response

  • List agents (list_agent) to list available agents with their ID, project IDs, and project names

Enable Preview Feature: MCP Access for Mosaic Models

Only needed for Strategy One (February 2026). The preview feature is available out of the box beginning in Strategy One (March 2026).

In the Collaboration Service's config.json file, set uslEnabled to true.

Copy 
"mcpConfig": {
  "enabled": true,
  "uslEnabled": true,
 }