Hey, good question. The multi-instance approach you outlined will work, but there’s one nuance worth knowing about.
Context switching: Yes, Cursor treats these as separate MCP servers with unique names (figma-backoffice, figma-frontoffice). You can tell the agent to use a specific one. That said, since both instances of @modelcontextprotocol/server-figma expose the same underlying tool names, there used to be a routing bug where Cursor would send all calls to whichever server registered first. This was fixed in Cursor 2.3 (related thread), but at least one user reported intermittent issues afterward. Make sure you’re on the latest version.
Resource usage: Running multiple stdio-based MCP instances has minimal overhead. Each one is just a separate Node process. No conflicts expected at that level.
A couple of tips:
-
If you hit routing issues (the agent calls the wrong server), the proven workaround is to fork the MCP server code and give each instance unique tool names. This thread describes the approach.
-
Figma now has an official SSE-based MCP server at
https://mcp.figma.com/mcp(listed in our docs). It might be worth checking if it fits your multi-project workflow better than the npm package. -
Being explicit in your prompts helps. Something like “Use the figma-backoffice MCP to get the layout for page X” will guide the agent to the right server.
Let me know if you run into any issues with the setup.