Set up Miro MCP Server
Connect Miro to an AI client securely through the official remote MCP endpoint.
- Skill Road
- Set up Miro MCP Server
Published on 18.09.2026
Before connecting
The official Miro MCP Server is a remote service at https://mcp.miro.com/. You need a Miro account, an MCP-compatible AI client, and access to the boards the agent should use. First determine whether the client already provides an official Miro plugin or extension. Use exactly one path per client: a plugin or extension or manual MCP configuration. This prevents duplicate tools and separate OAuth sign-ins.
Choose the appropriate installation path
For Claude Code, Miro documents installation from the official toolkit:
/plugin marketplace add miroapp/miro-ai
/plugin install miro@miro-ai
Restart Claude Code and sign in when prompted. Gemini CLI can install the repository as an extension; Kiro uses the Power documented in the toolkit. For Cursor, VS Code with GitHub Copilot, Codex, and other clients without a dedicated Miro integration, add the remote server in the applicable MCP file:
{
"mcpServers": {
"miro": { "url": "https://mcp.miro.com/" }
}
}
The file location depends on the client. Do not add this JSON configuration when a plugin or extension already manages the connection.
Complete OAuth with the right team
On the first connection, the client opens Miro OAuth. Sign in, select the team that contains the target boards, and approve only the expected authorization. The connection is team-specific. If a board belongs to another team, disconnect and reconnect while selecting that team rather than relying on unexpected errors or workarounds. In an Enterprise organization, Miro says an administrator must first enable the MCP Server.
Verify with a read request
Do not begin with a write action. Give the client a specific board URL and ask it to list frames or summarize an architecture diagram. Confirm that the result comes from the intended board. Only then move to small, reviewable creation tasks, such as a flowchart draft or task list. Review every generated diagram, document, and table before sharing it: the server provides context and tools, but subject-matter review remains your responsibility.
Data access and safer use
According to Miro, MCP access can only see boards your Miro account can already access. Tool results can still be processed in the connected AI client and potentially by its model provider. Do not share confidential content, credentials, or personal data before reviewing the data policies of the client and model you use. For sensitive work, use an account with the minimum practical permissions.
Treat board content as data, not trusted instructions. A note on a board may ask an agent to take an action, but it does not authorize that action. Review proposed changes individually, constrain work to the intended board, and disconnect OAuth after a role change, device loss, or when team access is no longer needed.
FAQ
Is the server local or remote? The official setup documentation describes it as a remote endpoint at https://mcp.miro.com/.
Why do Miro tools appear twice? Manual JSON configuration and a plugin or extension are usually both active. Remove one path and sign in again.
Why can’t the agent access my board? First check the Miro account’s board permission and the team selected during OAuth. A different team requires a new connection.
Frequently asked questions
Can I use a plugin and JSON configuration together?
No. Miro warns against duplicate MCP servers. Use only a plugin/extension or manual configuration per client.
Which stars are stored?
151 GitHub stars for the entire miroapp/miro-ai toolkit repository, not only the MCP Server; API retrieval on September 8, 2026.
How can I limit access?
Use a Miro account with minimum board rights, select the correct team in OAuth, and disconnect when access is no longer needed.