Canva Dev MCP Server
Official local MCP server by Canva for developers: Canva documentation, App SDK, and API reference available directly in your AI coding client.
- Skill Road
- Canva Dev MCP Server
Categories
Description
The Canva Dev MCP Server is Canva's official Model Context Protocol server for developers building Canva Apps or Connect API integrations. It runs locally on the development machine and fetches documentation from canva.dev and related sources on demand. This gives a compatible AI client — such as Cursor, Claude Desktop, or Claude Code — specialized tools and up-to-date documentation context without the developer needing to switch between the assistant and a browser. The official documentation is available at https://www.canva.dev/docs/apps/mcp-server/.
What is the Canva Dev MCP Server?
The server is not a design automation tool and has nothing in common with the Canva AI Connector (the remote MCP endpoint at mcp.canva.com/mcp). It is a pure developer assistant: it brings Canva developer documentation, the App UI Kit, the Apps SDK, and the Connect API reference into the AI agent's context, so that the agent can provide more precise, documentation-backed answers when writing Canva App code. According to the provider, the server supports AI clients such as Cursor, Claude Desktop, and Claude Code.
Distinction from the Canva AI Connector
Canva publishes two separate MCP products with different purposes. The AI Connector (Canva MCP Server) is a remote server at https://mcp.canva.com/mcp that connects via OAuth to a Canva user account and enables design actions such as creating, editing, searching, and exporting designs. It is aimed at content and marketing teams. The Canva Dev MCP Server, by contrast, is a local stdio process launched by the Canva CLI that provides exclusively read access to public developer documentation. It is aimed at developers. Both servers come from Canva but address different target audiences and use cases — one for design work, the other for app development.
How does the server work?
The Canva Dev MCP Server is started via the Canva CLI (@canva/cli) and configured as a local MCP endpoint in the respective client. For Cursor, a .cursor/mcp.json file is created containing the start command npx -y @canva/cli@latest mcp. Claude Desktop and Claude Code receive analogous configuration files. According to the documentation, MCP tools are only available in the client's Agent mode and cannot be invoked directly: the AI agent decides which tool to use based on the request. To activate the server, a context-relevant question with keywords such as "App UI Kit," "Apps SDK," or an explicit request for Canva documentation references is sufficient.
Requirements
For operation, git, Node.js v24 or higher, npm, and a compatible MCP client are required. According to the official documentation, Cursor, Claude Desktop, and Claude Code are explicitly named. An internet connection is required so that the server can load documentation from canva.dev on demand. A Canva account or API key is not needed, since the server accesses only public developer documentation.
Step-by-step setup
Setup for Cursor involves three steps: first, create the configuration file .cursor/mcp.json in the project directory. Second, add the entry { "mcpServers": { "canva-dev": { "command": "npx", "args": ["-y", "@canva/cli@latest", "mcp"] } } }. Third, restart Cursor to load the settings. The procedure is analogous for Claude Desktop and Claude Code: the respective MCP configuration file receives the same start command. Canva also provides a direct Cursor configuration link that skips manual file creation. After restarting the client, verify the connection by asking a question in Agent mode — for example, "How many components are in the App UI Kit?" — and wait for a tool invocation prompt.
Features and tools
According to the provider, the server gives an AI agent access to specialized tools and documentation context for Canva App development. Specifically, the assistant enables: documentation search on canva.dev for Apps, the App UI Kit, the Apps SDK, the Connect API, and related topics; API reference queries about available components, hooks, and configuration options; context-relevant answers about app configuration, authentication, scopes, monetization, and the submission process; and code suggestions based on current Canva design guidelines and component definitions. The server has no public source repository; Canva distributes the package through the Canva CLI.
Security and privacy
According to the official documentation, the Canva Dev MCP Server operates locally on the device and does not transmit code or prompts to any sources other than the configured AI agent or the language model in use. It fetches documentation from canva.dev and other sources and provides it to the agent as context. Developers should be aware that retrieved documentation content is embedded in the AI client's model requests. When working with proprietary model backends, it is worth verifying what data the requests contain. Theoretical prompt injection through manipulated documentation content is practically very unlikely with official Canva documentation. Since the server requires no Canva account and no authentication, there is no risk of accidental design changes or data access.
FAQ
Is this server different from the Canva AI Connector? Yes. The Canva AI Connector is a remote MCP server for design work that connects via OAuth to a Canva account. The Canva Dev MCP Server is a local stdio process for developers that fetches public Canva developer documentation. Both come from Canva but address different audiences.
Do I need a Canva account? No. The Dev MCP Server accesses only public documentation and requires neither an account nor an API key.
Is the server free? The Canva CLI package @canva/cli is free to use. Costs may arise from the AI client or the language model in use.
Is there a public repository? According to research conducted on 2026-09-08, Canva does not publish standalone source code for this MCP server; it is distributed through the Canva CLI.
Requirements
git, Node.js v24 or higher, npm, compatible AI client (Cursor, Claude Desktop, Claude Code), internet connection for documentation fetching.
Installation instructions
Quick start
Add to .cursor/mcp.json (or client equivalent): { "mcpServers": { "canva-dev": { "command": "npx", "args": ["-y", "@canva/cli@latest", "mcp"] } } }. Restart the client.
What the Canva Dev MCP Server is
According to the provider, the Canva Dev MCP Server is an official Canva tool that provides AI-powered development assistance for Canva apps and integrations. It is built on the Model Context Protocol, an open standard that gives AI applications structured access to external tools and documentation. Connecting a compatible MCP client such as Cursor or Claude Desktop to the server at canva.dev gives the attached AI model access to specialized tools and documentation content tailored specifically to building Canva apps. It's worth distinguishing this from a separate Canva offering, the Canva AI Connector, which helps end users create actual designs. The Dev MCP Server, by contrast, targets developers building apps and integrations for the Canva platform.
Step 1: Prerequisites
Per the official documentation, using the server requires Git, Node.js version 24 or later, and npm. A compatible MCP client is also needed; examples given include Cursor, Claude Desktop, Claude Code, and Visual Studio Code. Anyone who hasn't set these tools up yet is pointed to a separate guide covering the general prerequisites for Canva apps, which describes the basic development environment for the Canva Apps SDK. The MCP server itself doesn't replace these fundamentals; it assumes a working development environment already exists and serves as an additional knowledge and tool source for the AI model within it.
Step 2: Setup depending on the MCP client
The concrete configuration differs by client but follows the same underlying principle: the client gets an entry that launches the server via the npx command using the Canva CLI package and the mcp subcommand. For Cursor, this configuration can be set up automatically through a prepared configuration link, or created manually as a file inside the project directory. After saving the configuration, the respective MCP client needs to be restarted for the changes to take effect. Whether the connection succeeded can then be checked through visual indicators in the client, for example by managing connected servers in Claude Desktop, or by asking the assistant a simple test question about the Canva App UI Kit and watching whether a matching tool call gets proposed.
Security, practical value, and limits
According to the provider, the MCP server runs locally on the developer's own device and fetches documentation from canva.dev and other sources; a user's own code or prompts are transmitted solely to the AI model or tool being used and are not additionally forwarded anywhere else. This architecture reduces the risk of sensitive source code leaking to third parties unintentionally, but it doesn't replace the usual care needed when handling API keys or credentials within one's own development environment. In practice, the server mainly helps find current information about App UI Kit components, Apps SDK interfaces, or best practices faster and more precisely than a general web search would. At the same time, its usefulness depends on the quality of the questions asked: the documentation recommends using specific keywords such as App UI Kit or Apps SDK and starting a new session whenever answers seem unclear or inconsistent, since context errors can otherwise persist across multiple messages. Anyone who isn't building Canva apps but simply wants to create designs is better served by the separate Canva AI Connector rather than this developer tool.
npx -y @canva/cli@latest mcp
Authentication
No authentication required; the server accesses public Canva developer documentation.
Required access permissions
Read-only access to public Canva developer documentation; no account or design access.
Transmitted or stored data
The server runs locally and fetches documentation from canva.dev. Retrieved content is embedded in the AI client's model requests. According to the provider, no code or prompt is transmitted to other sources.
Security risks
Low risk profile: read-only access to public documentation, no authentication, local stdio process. Retrieved documentation flows into model requests.
License and costs
- License
- Not recorded yet.
- Cost
- free
The Canva CLI package @canva/cli is free to use. Costs may arise from the AI client or language model. Current information at canva.dev.
Alternatives
Not recorded yet.
At a glance
- Provider
- Canva
- Status
- Official server
- Deployment
- Local
- Current version
- Not recorded yet.
- Last reviewed
- 08.09.2026
Repository and documentation
Categories
Supported clients
Related guides
Guides and background related to this entry.