Set up Canva MCP Server

How the official Canva Dev MCP Server helps developers with Apps SDK context, running locally on your machine without touching actual design content.

Published on 09.09.2026

What the Canva Dev MCP Server Is

The Canva Dev MCP Server is a tool provided by Canva itself that offers AI-powered development assistance for Canva apps and integrations. According to the provider, connecting a preferred MCP-capable client such as Cursor or Claude Desktop to canva.dev grants access to specialized tools and documentation that enhance one's own development workflow. An important distinction made explicitly in the official guide is that this MCP server should not be confused with the separate Canva AI Connector, which helps end users with their designs inside the Canva editor itself. The dev MCP server described here instead targets exclusively developers building apps for the Canva platform, and it primarily serves as a documentation assistant for the Apps SDK and the App UI Kit.

Prerequisites for Setup

According to the official documentation, several base prerequisites are needed before using the server: Git must be installed, along with Node.js version 24 or later and npm as the package manager. A compatible MCP client is also required; the documentation names Cursor and Claude Desktop as examples but also supports Claude Code and Visual Studio Code. These tools need to be in place before the actual setup, since the server itself does not ship with any of these dependencies but is fetched on start via npx.

Setup Per Client

The actual setup step differs slightly depending on the tool used but follows the same underlying principle everywhere: the server is started via the npx command with the at-canva-slash-cli package at its latest version and the mcp subcommand. For Cursor, a configuration file is created in the project directory; for Claude Desktop, the same configuration is entered in the developer settings; for Claude Code, a single-line command via the claude mcp add function is enough; and for Visual Studio Code, a matching configuration file is placed in the workspace. In every case the server runs as a local process communicating over standard input and output, so it is not an externally hosted service with which one's own project data would need to be shared.

Security and Data Handling

A central security aspect lies exactly in this local execution model: the dev MCP server runs as a local stdio process on the developer's own machine and, according to independent descriptions from the MCP ecosystem, serves purely as a documentation assistant for development questions around the Apps SDK, without ever touching or modifying actual design content inside Canva itself. Anyone using the server in production should nevertheless, as with any locally run npx-based tool, make sure to always fetch the current official version and verify the package source, to avoid pulling in compromised dependencies.

Practical Value for App Developers

In everyday use, the server mainly helps by providing direct access, right inside the editor, to current SDK references, design guidelines, and sample code while developing a Canva app, without needing to switch context and search the online documentation manually. This considerably speeds up typical tasks such as correctly integrating elements, configuring permission scopes, or understanding authentication flows within a Canva app, because the AI assistant works with tailored context pulled from the official documentation rather than relying on potentially outdated training knowledge.

Limits and Positioning

The distinction remains important: anyone who only wants to create finished designs in Canva or edit them with an AI assistant does not need this developer server and should instead use the separate Canva AI Connector referenced in the Canva Help Center. The server described here also does not bring any design or publishing capabilities of its own, only development assistance, which is why it is only meaningfully usable in the context of an active app development project.

Published on 09.09.2026

Categories

Frequently asked questions

Do I need a paid Canva account?

Not for the core features. Design generation, editing, search, export, comments, and asset uploads work on all plans according to Canva, including the free one. Resizing and brand-template search require Canva Pro, and autofill with your own content is reserved for Enterprise accounts. generate-design additionally draws on the account credit model.

Does the server run locally?

No. Canva operates it exclusively as a hosted service at https://mcp.canva.com/mcp. Clients without native remote support connect through the npx -y mcp-remote@latest bridge. There is no public repository and no self-hosted variant.

Which AI assistants are supported?

Canva's documentation lists setup guides for Claude, ChatGPT, Gemini, Cursor, and VS Code. In principle any MCP client that can connect to a remote server via OAuth will work.

What happens to my data?

Each connection is authorized per user via OAuth, and agents act only within the signed-in account's permission scope. Canva's usage policy prohibits using Canva-sourced data, prompts, outputs, or brand kit elements to train AI models; Canva may withdraw MCP access for violations.

Can the assistant change existing designs?

Yes, through editing transactions (start-editing-transaction, perform-editing-operations, commit-editing-transaction). Such write actions should be reviewed before they run; a dedicated folder for AI drafts helps avoid accidental changes to finished material.