Connect a Client Using an MCP Server

Interact with agents from multiple clients and with Mosaic models in a ChatGPT-like interface connected to an MCP Server. Before you can interact, you must connect the client.

For an overview of using a MCP server, see Interact with Agents, Mosaic Models, and Cubes Using an MCP Server.

Connecting a client is the same procedure for agents and for Mosaic; only the URL differs.

Copy the URL for your server from Server URLs and add it in your client, then sign in to Strategy Library when prompted. Connect to Clients provides steps for common MCP clients.

Prerequisites

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 certified and added to the user's Library. Certify only the agents you want to expose.

  • Mosaic models must be certified to be accessible through MCP. 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.

Server URLs

In the following URLs, env-domain serves as a placeholder for your environment ID. Replace env-domain with your environment's domain and append the corresponding URL.

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

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

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

Connect to Clients

Enable the MCP Server before connecting MCP clients to it for interaction with agents and Mosaic.

Select the connection procedure for the client you are working with:

Connect to the MCP Server using Claude

Who sets up the connection depends on the plan.

  • On Team and Enterprise, an Owner registers the server once for the whole organization. Users then authenticate to it separately, so no one inherits access they did not already have in Strategy.

  • Pro and Max do not have an organization layer, so each user does both registration and authentication.

Team and Enterprise plans: Owner

  1. Open Claude.

  2. From the Organization settings menu, select Connectors and click Add.

  3. Point at Custom and choose Web.

  4. Type your server URL using the following syntax:

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

    Replace domain with your MCP server.

  5. Click Add.

    Do not change the Advanced settings. Strategy issues the client registration itself, so the Client ID and Client Secret fields stay empty.

Team and Enterprise plans: Users

  1. Open Claude.

  2. From the Customize menu, select Connectors.

  3. Locate the entry that your Owner set up, which is tagged Custom.

  4. Click Connect next to the connector.

  5. Strategy Library opens. Log in with your credentials.

Pro and Max plans

  1. Open Claude.

  2. From the Customize menu, select Connectors.

  3. Click + and choose Add custom connector.

  4. Type your server URL using the following syntax:

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

    Replace domain with your MCP server.

  5. Click Add.

  6. You can skip the Advanced settings.

  7. Click Connect next to your new connector.

  8. Strategy Library opens. Log in with your credentials.

Connectors are switched on per conversation rather than globally. In a chat, click + at the lower left, open Connectors, and toggle on your connector.

Once you create a connector, you cannot edit it. Changing the URL means removing the connector and setting it up again.

For more details and the most up-to-date steps, visit the vendor documentation: Anthropic: Get started with custom connectors using remote MCP.

Connect to the MCP Server using Copilot Studio

  1. If you are using Strategy (July 2026) or later, complete the steps below once for each environment. If you are using an earlier version, use manual OAuth instead of following the steps.

    1. Open Workstation.

    2. In the Navigation pane, click Environment Settings.

    3. In the Environment tab, click Library.

    4. In the Model Context Protocol for Agent section, add https://global.consent.azure-apim.net/redirect/*, including the wildcard *, in the Callback URL box.

    5. Click Save.

  2. Open Copilot Studio.

  3. Click Agents in the toolbar on the left.

  4. Create a new agent by clicking Create Blank Agent in the toolbar or click an existing agent.

  5. Scroll down to the Tools section and click Add Tool.

  6. Click New Tool on the right, below the suggestions.

  7. Click Model Context Protocol.

  8. Type the Server Name, Server Description, and Server URL including the complete server path. For the URL, use the following syntax:

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

    Replace domain with your MCP server.

  9. For the Authentication, select OAuth 2.0.

  10. If you are using Strategy (July 2026) or late:

    1. For the Type, select Dynamic Discovery.

    2. Click Create.

  11. If you are using an earlier version, complete the steps below:

    1. For the Type, select Manual.

    2. Type your Client ID and Client Secret.

      • Access the Client ID and Client Secret in Workstation in the environment's properties. In the Navigation pane of Workstation, click Environment Properties. In the Environment tab, click Library. Scroll to the Model Context Protocol for Agent section. The Client Secret may only be available to your environment administrator.

      • Do not use the tenant code as the Client ID. An authentication error in Copilot Studio and Power Automate will result.

    3. Type the Authorization URL, Token URL Template, Refresh URL, and Scopes. See Server URLs for the values.

    4. Click Create.

    5. Copy the Redirect URL displayed in the Preview window. Provide it to the MCP owner, so they can help to add the redirect_url in the MCP configuration.

    • The MCP owner adds the URL and types offline_access in the Scopes box in Workstation, in the environment's properties. See above for steps to locate the Model Context Protocol for Agent section.

  12. Click Next.

  13. To establish the connection, click Not Connected.

  14. After you are signed in and connected, click Create.

  15. A new browser window displays. Strategy Library opens. Log in with your credentials.

You can now interact with your agent or Mosaic model using Copilot Studio.

For more details and the most up-to-date steps, visit the vendor documentation: Microsoft: Connect your agent to an existing MCP server.

Connect to the MCP Server using Gemini Enterprise

Create the Data Store

  1. Open your project in Gemini Enterprise and choose your App to create the agent. Select Create a data store.

  2. In Source, scroll down to third-party sources and select Custom MCP Server.

  3. In Data, enter the MCP Server URL using the following syntax:

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

    Replace domain with your MCP server.

  4. Type the Authorization URL, Token URL, and Scopes for the environment. See Server URLs for the values.

  5. Type your Client ID and Client Secret for the environment.

    • Access the Client ID and Client Secret in Workstation in the environment's properties. In the Navigation pane of Workstation, click Environment Properties. In the Environment tab, click Library. Scroll to the Model Context Protocol for Agent section. The Client Secret may only be available to your environment administrator.

  6. Click Verify Auth. Once you are logged in, click Continue.

  7. Skip the Advanced options.

  8. In Configuration, set your MCP connector name and click Create. It takes a few minutes for the connector to be created.

  9. When the connector Status changes from Creating to Active, click your newly created MCP connector name and confirm:

    • Connected apps shows your Gemini Enterprise app correctly

    • Connector State is Active

    • Action State is Active

  10. If Action State is not visible, create OAuth credentials:

    1. In the menu bar at the top left, from the API & Services menu, select Credentials.

    2. Select Create credentials at the top, then OAuth client ID.

    3. For the Application Type, choose Web application. Use the same name as your MCP connector.

    4. Add these Authorized redirect URIs:

      https://vertexaisearch.cloud.google.com/oauth-redirect

      https://enterprisesearch.cloud.google.com/oauth-redirect

      https://console.cloud.google.com/oauth

    5. Click Create.

    6. Optionally, to save the credentials, click Download JSON, then click OK.

    7. Verify that Action State is now Active.

  11. In the data store, select Actions in the left panel.

  12. Click Reload custom actions, select all the actions, and click Enable actions. Confirm the status changes to Enabled.

Create Your Agent Using the Data Store

  1. In Gemini Enterprise, open a new chat and authorize your MCP connector. Make sure it toggles On.

  2. In the left panel below Agents, click New Agent. Select Proceed to Builder.

    The Agent Designer opens.

  3. Click the Agent block to begin designing, then give your agent a Name, Description, Instructions, and Model.

  4. Under Connectors, toggle on the custom MCP connector that you created.

  5. Click Create. You can now chat with your agent.

You can now interact with your agent using Gemini Enterprise.

Connect to your Mosaic model Using the Data Store

For Mosaic MCP, use the system instructions below as a starting point. Modify them to suit your needs.

Copy

# System Instructions: Mosaic Data Assistant

# System Instructions: Mosaic Data Assistant

You are an expert data assistant for Mosaic, Strategy Software's governed semantic
layer. Your role is to translate natural language questions into accurate SQL
queries against Mosaic models by strictly following the tool sequence and syntax
rules below.

## 1. Runtime Tool Mapping & Execution Workflow

In this environment, your tools are dynamically prefixed with an internal hash
identifier (e.g., `custom_mcp_custom_mcp_[ID]_...`). You must identify and execute
the tools based on their functional endings.

To answer any data question, follow this exact sequence:

1. **Identify Project Workspace:** Execute the tool whose name ends in
   `get_projects` (dynamically matched as `${tool_prefix}get_projects`).
2. **List Models:** Execute the tool whose name ends in `get_models`
   (dynamically matched as `${tool_prefix}get_models`).
3. **Fetch Semantics:** Execute the tool whose name ends in `get_semantics`
   (dynamically matched as `${tool_prefix}get_semantics`) to retrieve attributes
   (dimensions) and metrics (governed measures). *Analyze these semantics before
   writing any SQL.*
4. **Execute Query:** Execute the tool whose name ends in `query` (dynamically
   matched as `${tool_prefix}query`) to run a read-only Trino-compatible SQL query.

## 2. SQL Syntax & Query Conventions

* **Identifier Quoting:** Attribute columns use a `<form>__<name>` pattern
  (e.g., `"customer name__customer name"`, `"campaign__campaign id"`). You **must**
  wrap these column names in **double quotes**. Backticks are strictly unsupported
  and will cause syntax errors.
* **Time Bucketing:** Date and timestamp attributes are exposed both as a base
  column and as pre-bucketed interval columns: `(day interval)`, `(month interval)`,
  `(quarter interval)`, and `(year interval)`.
  * *Rule:* Always use these interval columns for time-grouped aggregations. **Do
    not** use `date_trunc` or manual date math.

## 3. Governance & Accuracy Guardrails

* **Metrics over Raw Aggregations:** Strongly prefer Mosaic-defined metrics
  (returned via the tool ending in `get_semantics`) over raw column aggregations.
  Metrics encode validated business logic. Ad-hoc `SUM` or `AVG` over raw columns
  will frequently disagree with your reports.
* **Read-Only Enforcement:** The query tool is strictly read-only. Any write or DDL
  statements (`INSERT`, `UPDATE`, `DELETE`, `MERGE`, `CREATE`, `DROP`) will fail and
  return `MOSAIC_SQL_WRITE_REJECTED`.

## 4. Response Strategy Checklist

Before executing a query, verify:

1. Have I mapped my tool action to the correct dynamically prefixed tool name?
2. Have I found the exact pre-defined metric requested by the user?
3. Are all `<form>__<name>` columns explicitly double-quoted?
4. Am I using pre-bucketed interval columns for any required time groupings?

 

For more details and the most up-to-date steps, visit the vendor documentation: Google: Set up your custom MCP server data store.

Connect to the MCP Server using ChatGPT

  1. If you are using Strategy (July 2026) or later, complete the steps below once for each environment. If you are using an earlier version, use manual OAuth instead of following the steps.

    1. Open Workstation.

    2. In the Navigation pane, click Environment Settings.

    3. In the Environment tab, click Library.

    4. In the Model Context Protocol for Agent section, add https://chatgpt.com/connector/oauth*, including the wildcard *, in the Callback URL box.

    5. Click Save.

  2. Open ChatGPT.

  3. In the bottom left, click your name.

  4. Click Settings.

  5. In the navigation pane on the left, click Plugins.

  6. Click Developer mode to enable it.

  7. In the left panel of the ChatGPT interface, click Plugins, then click the + to create a new plugin.

  8. Fill in the information to create the plugin.

    1. Add an Icon image file (optional).

    2. Type a Name and an optional Description.

    3. Type the MCP Server URL using the format https://env-domain/collaboration/mcp/agent, replacing domain with your MCP server.

  9. From the Authentication drop-down list, select OAuth.

  10. Click Advanced Settings.

  11. From the Registration Method drop-down list, select Dynamic Client Registration (DCR).

  12. In the New App panel, select the check box next to I understand and want to continue.

  13. Click Create.

You can now interact with your agent or Mosaic model in ChatGPT.

For more details and the most up-to-date steps, visit the vendor documentation: Open AI: Developer mode and MCP apps in ChatGPT.

Connect to the MCP Server using AWS/Amazon Quick Suite

  1. Log into the AWS access portal.

  2. Click AdministratorAccess.

  3. Search for Amazon Quick Suite and select it.

  4. In the navigation bar on the left, click Connectors.

  5. Click the Create for Your Team tab. Click Model Context Protocol.

  6. Click Create New.

  7. Type the Name and Description.

  8. In the MCP Server Endpoint box, copy and paste your MCP server URL using the following syntax:

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

    Replace domain with your MCP server.

  9. Click Next.

  10. From the OAuth Configuration drop-down list, select Default OAuth App.

  11. If metadata discovery fails, it falls back to user authentication. Follow the steps below: 

    1. Type your Client ID and Client Secret.

      • Access the Client ID and Client Secret in Workstation in the environment's properties. In the Navigation pane of Workstation, click Environment Properties. In the Environment tab, click Library. Scroll to the Model Context Protocol for Agent section. The Client Secret may only be available to your environment administrator.

    2. Type the Authorization URL and Token URL. See Server URLs for the values.

  12. Click Create and Continue.

  13. Review the list of enabled actions. Click Next.

  14. After publishing, click Try It to test the MCP, or continue with the next step to test.

    Start Testing

  15. In the navigation panel on the left, click Chat Agents.

  16. Click Create Chat Agent.

  17. If you want to specify the agent's persona, describe it, then click Generate. If not, click Skip.

  18. Type a name in the box at the top of window.

  19. In the Actions section at the bottom of the window, click Link Actions.

  20. Select MCP Agent, then click Link.

  21. At the top of the window, click Launch Chat Agent.

Once the Successfully Launched Chat Agent message displays, you can interact with your agent using AWS.

For more details and the most up-to-date steps, visit the vendor documentation: AWS: Model Context Protocol (MCP) integration.

Connect to the MCP Server using VS Code Copilot

  1. Open Copilot.

  2. Open a new window using Ctrl+Shift+P.

  3. Search for and select MCP: Add Server.

  4. Choose HTTP (HTTP or Server-Sent Events).

  5. Paste your MCP server URL using the following syntax:

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

    Replace domain with your MCP server.

  6. Type a Name for the server.

  7. Select Global.

  8. Click Allow authentication.

  9. Click Open to open the external website.

  10. Strategy Library opens. Log in with your credentials.

  11. In the window, click Open Visual Studio Code.

  12. Confirm that your mcp.json file shows the server.

  13. Click Start above the mcp line.

  14. Click Allow to connect.

  15. Ensure that the MCP connection is running, by checking the mcp.json file for the Running status.

  16. Set the Chat panel to Agent mode before asking questions.

For more details and the most up-to-date steps, visit the vendor documentation: Visual Studio Code: Add and manage MCP servers in VS Code.