How to set up Figma MCP: Remote and Desktop paths

Figma MCP lets an AI client work with context from a Figma design. What it can read or change depends on the client, file permissions and workspace settings. Start by connecting the client to the file you want to use.

Quick answer: start with Figma's Remote MCP path when your client supports it. Use a Desktop MCP path only when the official client instructions or your organization require it. Do not paste a token into an unknown third-party configuration.

1. Remote MCP: the recommended starting point

1. Open Figma's official MCP Server documentation and confirm that your client is listed or supported.
2. Sign in with the Figma account that can open the target file. A link alone does not grant edit permission.
3. In the MCP client, use its Add/Connect flow and complete the Figma authorization screen. Review the requested scopes before approving.
4. Start with a read-only request such as “summarize the selected frame and list its components.” Check the file name, page and selection before asking for any change.
5. For an edit, ask for one small, reversible change and review the result in Figma. The client may require an explicit approval for each write operation.

2. Desktop MCP: use it only when required

Some teams still use a desktop client or an organization-specific connector. Follow the current instructions for that client rather than copying a configuration from an older blog post. The desktop route can require a local runtime, an approved client and an access token; the required fields and token lifetime are client-specific. Store secrets in the client’s secret store or environment, never in a public repository or screenshot.

3. MCP and API are different

• API: your program calls a documented endpoint with a defined request and response.
• MCP: an MCP host exposes tools and context to an AI client. The model still acts within the permissions and approval rules of the connected service.

MCP does not automatically grant access to every page, component or team library. A permission error is usually fixed in Figma's file/team access settings, not by adding a second secret.

4. A safe design-to-code workflow

1. Duplicate or branch the working file before testing a write.
2. Ask the client to identify the file, page and frame before changing anything.
3. Request a small change with an explicit acceptance criterion, for example “change the button radius to 8 px and leave the label unchanged.”
4. Review layout, typography, variables, components and prototype links in Figma.
5. Record the client version, date, file permission and result so another teammate can reproduce it.

5. Troubleshooting

The Figma tool is not listed

Check that the client is supported, update it, sign in again and confirm the MCP connection is enabled for the current workspace.

Read works but edit fails

Check file edit permission and the client's write approval. Do not infer that a read-only connection can write.

The request is too large

Limit the selection and ask for one page or frame at a time. Large files can exceed a client's context or request limit.

A secret was exposed

Revoke it in the issuing service, remove it from logs and rotate it before reconnecting.

6. Where Pixso fits

Pixso is a separate collaborative UI-design workspace. If your team needs to finish a UI, prototype or handoff after an AI experiment, open the Pixso workspace and verify the current plan and permissions in your account. Treat any Pixso MCP or AI connector as a separate product feature: check its current documentation and approval scope instead of assuming that Figma MCP settings transfer.

Sources

Primary source: Figma MCP Server documentation.