Allow ${workspaceFolder} in MCP project configration

Feature Requests
1

I’ve been loving the ability to use custom MCP servers in Cursor, especially using the memory MCP (@itseasy21/mcp-knowledge-graph) to retain context across sessions. However, my team has encountered an issue with the configuration that’s making it difficult to deploy consistently across multiple developers’ environments.

Currently, we need to hardcode absolute paths to the memory file in our mcp.json config, like this:

{
  "mcpServers": {
    "memory": {
      "command": "npx",
      "args": ["-y", "@itseasy21/mcp-knowledge-graph"],
      "env": {
        "MEMORY_FILE_PATH": "/Users/myname/projects/my-project/memory.jsonl"
      }
    }
  }
}

This means each developer has to maintain their own version of the config file with their specific file path, which is not ideal for version control or onboarding new team members.

It would be incredibly helpful if Cursor supported the ${workspaceFolder} VS Code variable in the mcp.json configuration, similar to how VS Code handles this in launch.json and other config files. This would allow us to write a portable config:

{
  "mcpServers": {
    "memory": {
      "command": "npx",
      "args": ["-y", "@itseasy21/mcp-knowledge-graph"],
      "env": {
        "MEMORY_FILE_PATH": "${workspaceFolder}/memory.jsonl"
      }
    }
  }
}

This would make it much easier for teams to share consistent MCP configurations and would align with expectations for developers coming from VS Code. It would also make it simpler to set up memory-based MCPs that store their data within the project folder rather than in arbitrary locations.

Would love to see this feature added to improve the team development experience with custom MCPs. Thanks for considering!

6 Likes
2

hope to have this feature too. Not only for the project-specified knowledge, also for the task management like mcp server GitHub - cjo4m06/mcp-shrimp-task-manager: Shrimp Task Manager is a task tool built for AI Agents, emphasizing chain-of-thought, reflection, and style consistency. It converts natural language into structured dev tasks with dependency tracking and iterative refinement, enabling agent-like developer behavior in reasoning AI systems.

3

please this is very important to use global mcp server like file system

4

Much needed feature. I am also stuck due to this

5

And I want a cwd parameter to specify the working directory for the stdio transport.

6

I just figured out that you can use . and it will be expanded to the workspace root. Note that ./folder will NOT be expanded!

{
  "mcpServers": {
    "jupyter": {
      "command": "the-notebook-mcp",
      "args": [
        "start",
        "--allow-root",
        "."
      ]
    }
  }
}
7

I use everything-mcp to check the environment variable and found a variable named ‘WORKSPACE_FOLDER_PATHS’. This variable seems store the workspace path, but not sure this is a stable feture or not.

image
image689×363 52.9 KB