Detailed Analysis
A user on the Claude AI subreddit raises a practical interoperability question that surfaces a meaningful architectural gap in Anthropic's current product ecosystem: custom skills built within Claude Code cannot be straightforwardly ported to Claude Desktop or Cowork. The user reports having developed several skills through a Claude Code plugin and is seeking a supported method to expose those same skills — including slash commands — across Claude's other interfaces. Their attempt using `claude mcp serve`, a command that exposes file and shell tools via the Model Context Protocol, confirmed that skills and slash commands are not surfaced through that pathway, leaving no clean, documented bridge between the environments.
The question reflects a deeper structural reality about how Anthropic has architected its Claude product suite. Claude Code, Claude Desktop's Chat and Cowork modes, and the standalone Claude Code Desktop application are distinct environments with overlapping but non-identical feature sets. Skills — reusable, customizable prompt-and-workflow units triggered via slash commands like `/tdd` or `/write-a-prd` — are a compelling productivity layer that Anthropic has been actively developing and promoting. However, the research context makes clear that skills are currently scoped to specific interfaces rather than operating as a universal, cross-environment capability layer. This siloing is a notable friction point for power users who build sophisticated workflows in one context and reasonably expect portability across Anthropic's own first-party tools.
The Model Context Protocol (MCP) represents Anthropic's primary mechanism for extensibility, but as the user's experience illustrates, MCP in its current form is oriented toward exposing external tools and data sources — file systems, shell environments, APIs — rather than surfacing higher-order constructs like skills or slash command registries. This distinction matters: MCP is a tool-access layer, not a workflow-sharing layer. Bridging skills across Claude surfaces would require either a dedicated skills-sharing API, a centralized skill registry accessible by all Claude interfaces, or deeper MCP extensions that can represent and execute skill definitions. None of these appear to be officially supported as of April 2026.
The broader trend this question illustrates is the tension between Anthropic's rapid product expansion and the integration maturity that enterprise and power users demand. Anthropic has moved quickly to release differentiated products — Code for developers, Cowork for knowledge workers, Desktop for local agentic tasks — each with tailored capabilities. That differentiation is strategically coherent, but it creates interoperability debt. Communities like the one catalogued at the `awesome-claude-skills` GitHub repository are already building and curating skill libraries, signaling that users want a composable, portable skills ecosystem. Whether Anthropic responds with a formalized cross-surface skill portability mechanism will be a significant indicator of how seriously it intends to unify its product experience for the growing segment of users who operate fluidly across multiple Claude interfaces.
Read original article →