{"version":"2.1.312","anchor":"mcp-skills-hidden-from-the-deniallocal-display-path-mcp-pe","canonical_anchor":"mcp-skills-hidden-from-the-deniallocal-display-path-mcp-pe","heading":"MCP skills are handled separately, and MCP servers can have their own permission mode","tier":"notice","area":"MCP","scope":"both","heads_up":false,"url":"https:\/\/changelogs.core-directive.com\/v\/2.1.312\/e\/mcp-skills-hidden-from-the-deniallocal-display-path-mcp-pe","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.312","markdown":"### MCP skills are handled separately, and MCP servers can have their own permission mode\n\nSkills that come from MCP servers are now treated separately in permission handling, and each MCP server can have its own permission mode override\n\n**Unclear.** It is not clear whether `mcpPermissionModeOverrides` is something you can set yourself, or how it gets filled in.\n\n**What**\n\nMCP (Model Context Protocol) is a standard way to connect Claude Code to outside tools and services, each provided by an \"MCP server\". This release changes two things about how MCP is handled when Claude Code checks permissions:\n\n- Skills that come from an MCP server are now excluded from one path that handles and displays permission decisions locally.\n\n- Claude Code can look up a permission mode override for a specific MCP server by its name, stored as `mcpPermissionModeOverrides`.\n\n**Why**\n\nPermission handling for MCP skills and servers can now differ from that for other tools, and the permission mode can differ from one server to another.\n\n- Area: MCP\n- Tier: You'll notice\n- Useful: 3\/5\n- Signal: 3\/5\n- Scope: both\n- Heads-up: no"}