# MCP Servers Useful for Designers: What to Connect and What to Review > An MCP server connects an agent to your tools. What the Figma and Linear servers offer, which permissions they ask for and how to choose which to connect. *Tags: ai, mcp, tools, design* --- > [!info] Quick Definition > An **MCP server** is a program that exposes the tools, data or workflows of an application so an AI agent can use them through the MCP protocol. For a designer, the most useful ones connect the design file and the task board to the agent. ## What kind of servers is it worth connecting? An [[mcp|MCP server]] is worth connecting when the agent needs something that lives in another tool. Three categories are common in design work: - **Design:** access to files, components, variables and frames. - **Task management:** issues, projects and comments. - **Documentation and knowledge:** pages, databases and team notes. They do not all carry the same weight. Connecting a tool the agent does not need adds noise, and each server adds [[tool-use|tool definitions]] to the [[context-window|context]] the model reads. ## What does the Figma server offer? According to Figma's documentation, its MCP server gives agents the design context they need to generate code from your files. Today it can: - Generate code from selected frames. - Extract variables, components and layout data. - Use FigJam content and Figma Make resources as context. - Keep code aligned with your components through Code Connect. - Write native content to Figma Design, FigJam and Figma Slides, with the remote server only. There are two modes. The remote server, which Figma recommends, is hosted at `https://mcp.figma.com/mcp`, does not need the desktop app and has the broadest set of features. The desktop server runs locally with the app and is aimed at specific organization needs. According to the documentation, the desktop server requires a Dev or Full seat on the Professional, Education, Organization, Government or Enterprise plans, and writing to the canvas outside drafts requires a Full seat and the remote server. That feature is free during the beta and will move to usage-based pricing. Plans change, so confirm the current information before you decide. For the full workflow, see [[figma-to-code-with-mcp|from Figma to code with MCP]] and [[code-connect|Code Connect]]. ## What does the Linear server offer? According to Linear's documentation, its MCP server gives compatible models and agents access to your data, and its tools can find, create and update issues, projects and comments. It connects at `https://mcp.linear.app/mcp`. The interactive setup uses OAuth 2.1, and there are two ways to limit it to reading: the `https://mcp.linear.app/mcp/readonly` endpoint, or requesting only the `read` scope on the regular endpoint. A designer can use it so the agent reads the state of a flow and proposes tasks, while leaving the final creation to a person. More detail in [[linear-with-agents|Linear with agents]]. ## How do you choose which ones to connect? 1. **List the real work.** Which tasks do you repeat every week that require reading another tool? 2. **Review the capabilities.** Read the list of tools each server exposes. If you cannot find it, do not connect it yet. 3. **Start read-only.** When a read mode exists, use it. Move to writing only once the flow is tested. 4. **Decide local or remote.** A local server runs on your machine; a remote one is hosted by the provider and depends on its authorization. 5. **Review it periodically.** Tools and permissions change with server updates. Connecting a server is not a good idea when the data is sensitive and there is no clear access control, when you would use it only once (a manual query may be safer), or when nobody on the team will review what the agent can do through that connection. The MCP specification itself recommends that there always be a person able to deny tool calls, and that the application show which tools are exposed to the model. If your tool lets you, do not turn those confirmations off; more in [[action-approval|action approval]]. ## Example Illustrative example: a designer connects the Figma server so an agent can read the component library and generate the code for a form. Then she connects the Linear server with read-only permission, so the agent can mark which issues in the flow already have approved designs. Creating issues stays with her. Two connections, two permission levels, and a flow that can be audited. ## How is a server connected, and what happens when it changes? Each tool registers a server in its own way, but the principle is the same: you give a name, an address or the command that starts it, and the way to authorize. For a remote server with OAuth, the first connection opens an authorization window in the browser, and the tool stores the token. For a local one, the configuration points to a program on your machine. Before you save, check two things: that the address is the provider's official one and that the permission requested is the minimum. An address copied from a forum could point to a third-party server. Servers also change: a provider can add, rename or remove tools, and because the agent reads that list, what it does can shift without you editing anything. The MCP specification says a server that declares the `listChanged` capability should send a notification when its tool list changes. That notice depends on both the server and your tool implementing it, so review the capability list again every so often, especially before an important workflow. ## What can I do tomorrow? Review the MCP servers you have connected, if any. For each one, open its tool list and mark the ones that write. If you do not need some of them, disconnect the server or use the read-only version. Then note in the repository or the team documentation the list of servers, their permissions and who manages them. ## Common mistakes - Connecting everything available to "have more context." More tools means more noise and more risk. - Granting write permissions from the start. - Assuming the connection is as secure as the user's account without reading the configuration. - Not telling the team which servers are connected. ## Sources - [Figma MCP server, Figma documentation](https://developers.figma.com/docs/figma-mcp-server/) - [Linear MCP server, Linear documentation](https://linear.app/docs/mcp) - [What is the Model Context Protocol, official MCP documentation](https://modelcontextprotocol.io/docs/getting-started/intro) - [Tools, MCP specification (version 2025-11-25)](https://modelcontextprotocol.io/specification/2025-11-25/server/tools) --- Source: https://www.fernandoux.com/en/wiki/ai/mcp-servers-for-designers/