Improve Permission Handling in Filesystem MCP Server

Improve Permission Handling in Filesystem MCP Server

Issue

When using the Filesystem MCP server with Claude desktop app on macOS, there's a suboptimal user experience around system-level file permissions. Currently, permission prompts are triggered only after a failed file access attempt, leading to a jarring experience where:

1. User requests to read a file using read_file tool
2. Operation fails due to lack of permissions
3. Claude app prompts for system permissions
4. User has to retry the operation

Impact

This creates a poor user experience where:

• Operations fail before permissions are established
• Users see error messages before permission prompts
• Multiple permission prompts might appear for different files
• Operations need to be manually retried after granting permissions

Proposed Solution

Add a new check_permissions tool to the Filesystem MCP server that would:

• Allow proactive permission checking before file operations
• Trigger any needed system permission prompts upfront
• Return clear status about which paths are accessible
• Enable better error handling and user guidance

Implementation Details

• New tool: check_permissions
• Input: Array of paths to validate
• Output: Mapping of paths to their permission status
• Validates both MCP server allowed directories and system-level permissions
• Triggers OS permission prompts in a controlled manner

Benefits

• More predictable permission flow
• Better user experience
• Clearer feedback about access status
• Reduced need for error recovery
• Maintains separation of concerns within MCP architecture

Notes

While the ideal fix might involve changes to the Claude desktop app itself, this server-side solution provides a workable improvement within the current architecture.

Start by locating the Filesystem MCP server and its existing read_file operation, then inspect how allowed directories and system-level access are currently checked. Define the check_permissions tool around the stated array-of-paths input and path-status mapping, with completion requiring controlled permission prompts and clear access results.