# AI agents & MCP With Cards you have two options for using AI as agents. Either you create agents that can be used directly in the Cards interface, or you use our MCP connector to connect your own AI software (Mistral, Claude, or other compatible tools) to Cards. In the case of AI agents directly integrated into Cards, OpenAI is used as the engine. Using these agents consumes your AI credits, whereas MCPs do not (since your AI performs the processing). {% hint style="info" %} [Contact us](https://www.cards-microlearning.com/contact) to activate AI Agents and MCP integration. {% endhint %} ### List of AI agents and MCPs

The list of AI agents and MCP connectors is available on the "Apps & Integrations" page, then under "AI Agents & MCP Connectors." The tabs allow you to view the list of AI agents or MCP connectors. In the AI agent list, you will find two sections: one listing your agents (those you created) and the other listing agents created by other members of your team. ### What can an agent do? An AI agent has access to a list of tools, allowing it to interact directly with the content of your Cards space. These tools are the same as those used by MCP connectors. The tools are as follows: * Trainings Tool: allows access to, creation of, and editing of training courses * Chapters Tool: access, creation of, modification, and deletion of chapters * Cards Tool: access, creation, modification, and deletion of content cards * Categories Tool: access, creation, and editing of training categories * Routines Tool: access, creation, and editing of routines (and listing of routine categories) * Steps Tool: access, creation, editing, and deletion of routine steps * Giphy Search Tool: search for GIFs on Giphy * Unsplash Search Tool: search for images on Unsplash * YouTube Search Tool: search for videos on YouTube * Image Upload Tool: upload an image to your media library from a URL. Agents in Cards have access to all of these tools. For MCP connectors, you can generally choose whether or not to enable these tools in your AI software. For more granular permission control, you can create a custom role specific to your AI agent or connector. This permission control applies to Cards, so no connector or agent can override it. For more information, see the "[Custom Roles](/english/my-space/custom-roles.md)" documentation. ### AI Agents #### Create an agent

When you create an AI agent, you give it a first name, a last name, and a role. The content created by this agent will then be easily identifiable in your content listings (the "author" column will display the agent's first and last name). The role you choose must be a custom role. For more information, see the "Custom Roles" documentation. By assigning it a custom role, you can limit its actions on your content (for example, to only allow it to edit its own courses, and not those of other editors). Several configuration settings are available in the right-hand column: * Agent Type: allows you to choose the agent's behavior * Custom Instructions: allows you to "prompt" the agent by giving it instructions that will be applied when you call upon it * Availability: * Private: only you can use the agent * Public: the agent can be used by other editors in your space. #### Use an agent

To use an AI agent, click on the "AI" bubble in the bottom right corner. A panel will open on the side, allowing you to select the agent you want to use. Once you've selected an agent, a chat window will appear, and you can chat directly with the AI agent. In this chat, in the top right corner, you can view your previous conversations and pick up where you left off. The agent knows your first and last name, which allows it to know what content you've created. However, it currently cannot "see" which page you are on: you must therefore specify the name of the training course or its ID if you want the agent to interact with it. ### MCP Connector

To create a connector, you must, as with agents, choose a first name, last name, and role. Once the connector is created, a window will appear with your identification token. This token is only displayed once. Copy it; you will need it in your AI software (Mistral or Claude, for example). On the MCP edit page, you can regenerate a token if needed. The old token will be deleted and can no longer be used. The configuration of an MCP connector varies from one tool to another. Below are configuration examples for Mistral and Claude. Here is the information needed to connect the Cards MCP: * Name: Cards (the connector name is flexible, we recommend using "Cards") * Connector server/URL: * Authentication method/type: via an identification token (or "bearer token"). ### Configuration in Mistral To use MCP Cards in Mistral, go to The Chat ([https://chat.mistral.ai/chat](https://chat.mistral.ai/chathttps://chat.mistral.ai/chat)), then use the "Intelligence" > "Connectors" menu.

On this page, click the "Add a connector" button. A modal window will open allowing you to add a connector. Click the "Custom MCP Connector" tab. You can then enter the MCP information: * Name: Cards * Connector Server:

Mistral will then automatically detect the authentication method: * "API Token Authentication" * Header Name: Authorization * Header Type: Bearer You can now click "Create". Your connector will now appear in your listing. You can click on it to enter your authentication token:

Once the setup is complete, the MCP is ready to be used in The Chat. When you start a new conversation, you have the option to choose Cards as the tool. Click on the "+" in the bottom left corner, and you will be able to select Cards as the connector.

The Chat now has access to your Cards space. ### Configuration in Claude #### Prerequisites Since MCP Cards are not available in Anthropic's connector library, you will need to configure the MCP on your computer. Therefore, certain prerequisites are required for the connector to function. **Application version** To use MCP Cards in Claude, you must use the "application" version of Claude (and not the web version). **Node.js Libraries** The Cards connector uses token-based authentication. Therefore, the connector must be configured locally on Claude. To make it work, you will need to install Node.js on your computer. You can check if you already have Node.js installed by launching Claude, then going to Settings > Extensions, and then clicking on "Advanced Settings":

If Node.js is not present, you need to install it on your computer. {% hint style="danger" %} Note that on Windows, install Node.js in a folder without spaces. A bug in Claude occurs when the path to Node.js contains a space (such as "\Program Files\\"). {% endhint %} #### Configuration To add MCP Cards, you need to go to Settings, then "Developer".

On this page, click the "Edit config" button to open the folder containing the `claude_desktop_config.json` configuration file. Open this file and insert the following lines: ```json "mcpServers": { "cards-microlearning": { "command": "npx", "args": [ "mcp-remote", "https://api.cards-microlearning.com/mcp/cards", "--header", "Authorization: Bearer VOTRE_TOKEN_ICI" ] } } ``` Replace `YOUR_TOKEN_HERE` with your previously generated ID token from Cards. Your `claude_desktop_config.json` file may already contain lines, including your preferences. Therefore, you need to insert the lines above and remove your preferences. You should end up with something like this: ```json { "mcpServers": { "cards-microlearning": { "command": "npx", "args": [ "mcp-remote", "https://api.cards-microlearning.com/mcp/cards", "--header", "Authorization: Bearer 1234|xxxxxxxxx" ] } }, "preferences": { "chromeExtensionEnabled": false, "quickEntryShortcut": "off", "coworkScheduledTasksEnabled": true, "ccdScheduledTasksEnabled": true, "sidebarMode": "chat", "coworkWebSearchEnabled": true } } ``` Once the file is saved, you need to restart Claude. By going to "Connectors", the Cards connector will appear:

And in your Claude chat, you should now see the connector (enabled by default):

You can now ask Claude to interact with your Cards account. --- # Agent Instructions: Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter: ``` GET https://doc.cards-microlearning.com/english/apps-and-integrations/mcp.md?ask= ``` The question should be specific, self-contained, and written in natural language. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.