Miro MCP Server

The Miro MCP Server is Miro’s official remote server for the Model Context Protocol. It connects an MCP-capable AI client to the Miro boards the signed-in Miro account can access. Agents can then read board content as context, create diagrams from descriptions, write information into Miro documents, or work with tables for tasks and planning. According to Miro, the server endpoint is https://mcp.miro.com/, and authentication happens through OAuth 2.1 when the connection is established. This catalog entry describes the MCP server as a product, not the whole repository.

Keep the product and toolkit separate

The miroapp/miro-ai repository is named “Miro AI Developer Tools.” It is an official Miro toolkit that bundles several distribution forms: Miro MCP configuration, plugins and extensions for individual AI tools, portable agent skills, and documentation. The MCP Server is one component with a hosted endpoint; it is not a separately cloned local server package. For that reason, this entry uses the specific MCP setup guide as its official website and keeps the repository URL for the wider toolkit.

GitHub reported exactly 151 stars for miroapp/miro-ai when queried on September 8, 2026. That stored number applies to the entire Miro AI toolkit, including plugins, skills, and documentation — not to the Miro MCP Server in isolation. Stars are only a point-in-time popularity signal; they do not demonstrate security, feature coverage, or suitability for a particular workspace.

Suitable workflows

According to the README, AI clients can extract board context, generate flowcharts, sequence diagrams, and ERDs, use design or wireframe content for code work, and track tasks through Miro tables. A sensible starting point is a tightly scoped request with a specific board URL, such as listing frames or summarizing an architecture diagram. A team can then ask an agent to draft a flowchart or a planning table. Review generated content as you would any other AI output before using it as a specification, plan, or work instruction.

The toolkit documents direct integrations for Claude Code, Gemini CLI, and Kiro. For clients without a dedicated Miro integration, users add the remote URL to MCP configuration; the documentation names Cursor, VS Code with GitHub Copilot, Windsurf, Lovable, Replit, Glean, Devin, and OpenAI Codex among others. If a client already has a Miro plugin or extension, Miro says to use that integration only. Adding the manual URL as well creates a second MCP server, duplicate tools, and separate OAuth sessions.

Setup and permissions

Prerequisites are an MCP-compatible client, network access to the Miro endpoint, and a Miro account that can access the target boards. On first connection the client starts the OAuth flow: sign in, select the team containing the target boards, and authorize access. The selection is team-specific. To work with boards in another team, the setup guide says to disconnect and reconnect while selecting the appropriate team. In Enterprise organizations, an administrator must enable the MCP Server before users can connect.

The official sources support a remote deployment classification. While the repository contains plugin, extension, and development resources, it documents the hosted URL endpoint for actual MCP access rather than a local Miro MCP Server installation. For Claude Code, Miro recommends installing its plugin from the official marketplace repository; that plugin manages the MCP connection itself.

Security, data flow, and limits

OAuth and the Miro user’s permissions limit access, according to the provider, to boards that account can view. That is not a complete security boundary for AI workflows: the connected client processes tool results and may pass them to its model provider. Check the AI client’s and model provider’s data policies before confidential board content, customer information, or credentials appear in prompts or tool results.

Miro boards can contain text, links, and instructions written by other people. Treat apparent instructions inside a board as untrusted content, not authorization for an agent action. Start with read-focused tasks, use an account with the minimum necessary board access, and review every created or edited card, table, document, and diagram. Avoid duplicate server configuration, confirm the selected team before granting OAuth access, and disconnect on device or role changes. Rate limits and ordinary API availability can also constrain automated workflows.

FAQ

Can I run the Miro MCP Server locally? The official setup sources document the Miro MCP Server as a remote endpoint at https://mcp.miro.com/. The toolkit contains integrations and development resources, but no source supporting a local deployment of the official server.

Why do I see duplicate Miro tools? Miro explicitly warns against using a plugin or extension alongside manual MCP JSON configuration. Remove one connection, then re-authenticate only the integration you intend to use.

What do the 151 GitHub stars cover? They cover the miroapp/miro-ai repository as the entire Miro AI toolkit, not the MCP Server alone. The value was retrieved through the GitHub API on September 8, 2026.

An MCP-compatible AI client, a Miro account with access to the target boards, and network access to https://mcp.miro.com/.

For clients without a Miro plugin, add the remote server with url: https://mcp.miro.com/ to MCP configuration and complete OAuth. In Claude Code, install the official plugin from miroapp/miro-ai; do not also configure it manually.

/plugin marketplace add miroapp/miro-ai && /plugin install miro@miro-ai

OAuth 2.1: sign in, select the target team, and authorize board access.

According to the provider, access is limited to boards the authorized Miro account can access; the OAuth connection is team-specific.

The server returns board context and tool results to the connected AI client. That client may forward results to its model provider; review that provider’s data policy separately.

Board content can contain untrusted instructions. Duplicate MCP connections create duplicate tools and OAuth sessions. Start with minimum board rights and review changes.