Klaviyo MCP Server

The Klaviyo MCP Server connects AI assistants such as Claude or Cursor to the marketing platform Klaviyo. Anyone who maintains email and SMS campaigns, automated flows, and customer profiles there can give the assistant tasks in natural language, for example "How have our campaigns developed over the last month?" or "Create an email campaign for the new collection." The server is operated by Klaviyo itself. According to the documentation there is a hosted remote server, which is recommended, and a local server that runs on your own machine.

What is the Klaviyo MCP Server?

Klaviyo otherwise offers its functions through the web interface and a programming interface. The MCP server adds a third layer on top: it translates an assistant's requests into calls to the Klaviyo interfaces. According to the documentation the remote server is reachable at https://mcp.klaviyo.com/mcp and uses Streamable HTTP. The tools are organized by category in the documentation, which also notes which tools are read-only, which process user-generated content, and which are available on the remote server.

What is it particularly useful for?

The biggest gain lies in reporting and in creating drafts. The documentation names performance reports for campaigns and flows as well as creating email campaigns as examples. Marketing teams no longer have to click through several views; they phrase the question and check the result. Compared with a generic database server, this server knows the platform's concepts, such as profiles, lists, segments, campaigns, and flows, and works within Klaviyo permissions. For agencies that look after several accounts, the documentation also mentions that several Klaviyo accounts can be connected in the Claude connector.

Setup

With the remote server, sign-in works via OAuth with dynamic client registration. In Claude you add a custom connector with the server address under the connector settings and confirm the approval in the browser. In Cursor you enter the address as url in the mcp.json file and then sign in in the tool settings. In Visual Studio Code the configuration belongs in the MCP user configuration, which uses the key servers instead of mcpServers. The local server starts with uvx klaviyo-mcp-server@latest and needs a private Klaviyo API key in the PRIVATE_API_KEY environment variable.

Requirements, security, and control

For the remote server the documentation requires the Owner, Admin, or Manager role in the Klaviyo account. Particularly helpful is the fine-grained control through parameters on the server address: with read-only=true all tools that write data are disabled, with toolsets access can be limited to individual areas such as profiles or campaigns, and disable-tools-with-user-generated-content switches off tools that read user-generated content. The latter makes sense because such content could contain hidden instructions aimed at the assistant. With core-tools-only=true the tool list is limited to about 40 core tools, and beta=true unlocks new, possibly unstable tools. For the local server the variables READ_ONLY and ALLOW_USER_GENERATED_CONTENT play this role. According to the documentation the parameters cannot be set when the prebuilt Claude connector or the ChatGPT app is used.

Costs

The server requires a Klaviyo account. According to its pricing page there is a free entry plan with limits on active profiles and emails, alongside paid marketing plans that depend on profiles and message volume, and enterprise tiers through sales. For the MCP server itself the documentation names no separate fee. You can find the terms on the official pricing page.

Who is the Klaviyo MCP Server for?

It is worthwhile for e-commerce and marketing teams that already use Klaviyo and want to handle recurring reports or campaign drafts in conversation with an assistant. Without a Klaviyo account it is useless, because it is not a standalone marketing tool. Anyone who wants to start carefully begins with read-only mode and only enables write access once it is clear who may create campaigns through the assistant. The documentation makes no statement about the maturity of the server; new tools are only marked through the beta parameter. An open-source license for the package is not stated in the sources checked.

A Klaviyo account with the Owner, Admin, or Manager role (remote server) and an MCP client that supports remote servers, such as Claude or Cursor. The local server needs uv and a private API key.

With the Klaviyo MCP Server, an AI assistant analyzes campaigns and creates drafts. This guide leads from the prerequisites to the first query.

Step 1: Check the prerequisites

You need a Klaviyo account in which you are Owner, Admin, or Manager, and an MCP client that supports remote servers, such as Claude or Cursor. According to the documentation the server address is https://mcp.klaviyo.com/mcp.

Limit access up front

With your own client, append parameters to the address before connecting. read-only=true switches off all writing tools. With toolsets=profiles:read,campaigns:read you limit access to profiles and campaigns in read mode. disable-tools-with-user-generated-content=true disables tools that read user-generated content.

Step 2: Set up in Claude

Open the connectors in the settings, choose to add a custom connector, and enter a name and the address. Then confirm the sign-in and approval in the browser. Note that, according to the documentation, parameters cannot be set with the prebuilt Claude connector.

Step 3: Set up in Cursor

In the mcp.json file, add an entry klaviyo with the address as url. Cursor then shows "Needs login" in the tool settings; use this button to sign in to Klaviyo.

Step 4: Use the local server

Alternatively the server runs locally with uvx klaviyo-mcp-server@latest. For this you create a private API key with the required permissions in Klaviyo and enter it as PRIVATE_API_KEY. With READ_ONLY=true write access stays off here as well. Never store the key in the repository.

uvx klaviyo-mcp-server@latest

Step 5: First queries

Start with read tasks, such as "Summarize the performance of my latest campaigns" or "Which flows are active?". Spot-check numbers in Klaviyo. Only then is it worth allowing write access and letting the assistant create a campaign draft.

OAuth with dynamic client registration for the remote server; the local server uses a private Klaviyo API key in PRIVATE_API_KEY.

Read and write access within Klaviyo permissions; access can be restricted with read-only=true or toolsets.

The remote server runs hosted by Klaviyo. Retrieved customer and campaign data enters the AI client context and therefore reaches the model provider.

Campaigns created or changed unintentionally through write permissions, personal customer data passed to the model provider, hidden instructions in user-generated content, and an API key stored in the repository with the local server.