Skip to main content
One of the most compelling recent developments in the AI ecosystem is the Model Context Protocol (MCP). Briefly, MCP allows large language models (LLMs) like ChatGPT or Claude to perform actions in an app (called tools in the MCP protocol) on behalf of the user chatting with the LLM. MCP allows LLMs to increase their capabilities by connecting the AI to third party services in a seamless fashion. Stytch can be used to provide the authentication layer for implementing MCP servers, but, in addition, Stytch itself has been implemented as an MCP server. This allows configuring Stytch’s products using natural language through an MCP-capable AI — no need to visit the Stytch dashboard.

​
Configuring Stytch’s MCP server

Stytch is commonly configured during software development, so for this example we’ll be configuring an IDE with Stytch’s MCP server information. To one-click install Stytch’s MCP server with Cursor or VSCode, you can visit mcp.stytch.dev and click the button for installation. Alternatively, for other tools or to set up Stytch MCP with Cursor manually, proceed through the following walkthrough. Although the configuration process here is specific to Cursor, this example should give enough information to configure Stytch’s MCP server for other tools. In the app menu bar, activate 
Cursor => Settings => Cursor Settings
Select the Tools & Integrations tab, and then Add Custom MCP. This will open a file called mcp.json that contains the server’s configuration. Follow the documentation here to understand how to configure the JSON file. For our purposes, the file should look something like: 
{
  "mcpServers": {
    "stytch": {
      "url": "https://mcp.stytch.dev/mcp"
    }
  }
}
It may be necessary to restart Cursor to pick up the changes. Note: when configuring other tools, although the configuration steps may differ, the same URL used in the above configuration should be used to specify where Stytch’s MCP server is located. Now, when going to 
Cursor => Settings => Cursor Settings => Tools & Integrations
there should be an entry for stytch and a small message underneath which says “Needs authentication”.

​
Authenticating with the Stytch MCP server

From time to time it may be necessary to reauthenticate with the Stytch MCP server. In the Tools & Integrations view, you can click the message that says “Needs Authentication” to go through the auth flow to grant access to your Stytch workspace to Cursor. Once the flow is done, Cursor should enumerate the tools available and you should see a message like “21 tools enabled”. Clicking this message will expand to show you the tools that Stytch’s MCP server has implemented. Connected to Stytch MCP server The Stytch MCP server is now configured to be used with the Cursor agent. You should be able to ask it things like “What are my Stytch project IDs” and it will return a response. Using natural language—you don’t have to specifically match the name of the tool—you can perform any action listed in the available tools from Stytch’s MCP server. When running requests against the Stytch MCP server, note that, by default, Cursor is careful not to run commands automatically, as that may change state - you have to grant it permission for each tool call. If you get stuck or Cursor’s agent seems to hang, be sure to check if it’s waiting for permission to make the call. Cursor has documentation available on how to automatically run MCP tools if you choose to not be prompted for each call.

​
What’s possible using Stytch’s MCP server

Having access to Stytch via its MCP server is very handy for chatting with AI agents or quickly getting information without having to context switch out of the IDE. In addition, it’s convenient for keeping more of your systems in your repository; for instance, you can keep your email templates in git / version control and just ask the AI agent to upload a new version as it changes rather than having to log in and do the upload manually. One of the most exciting possibilities made available by using our MCP server is in putting configuration instructions directly in your repository’s README file. For an example of what this might look like you can check out our MCP example hosted on Vercel, which shows at the end of its README an example of how to script the configuration of Stytch via the MCP server and LLM. In this way you can create documentation that is both human-readable and which can be used by AI to bootstrap your projects.