Background
Peter Alexander (MCP Core Maintainer) suggested at the Feb 26 office hours that rather than having each server ship its own load_skill tool, clients should support model-driven resource loading directly — e.g., a built-in read_resource tool on the client side and a SDK-level list_skill_uris() method.
He argued it's a small lift for clients and would avoid duplicate approaches. This needs a proof of concept.
Scope
Build a minimal client (or client plugin/extension) that:
- Connects to one or more MCP servers
- Calls resources/list and identifies skill resources (by URI pattern, MIME type, or naming convention)
- Presents discovered skills to the user (list with name/description)
- On selection, calls resources/read to fetch the full skill content
- Injects the skill content into the model context
- Optionally subscribes to resources/subscribe for skill updates
Technology choice is open — could be a Python script using the MCP SDK, a VS Code extension, or a wrapper around an existing client.
Acceptance Criteria
References
Background
Peter Alexander (MCP Core Maintainer) suggested at the Feb 26 office hours that rather than having each server ship its own load_skill tool, clients should support model-driven resource loading directly — e.g., a built-in read_resource tool on the client side and a SDK-level list_skill_uris() method.
He argued it's a small lift for clients and would avoid duplicate approaches. This needs a proof of concept.
Scope
Build a minimal client (or client plugin/extension) that:
Technology choice is open — could be a Python script using the MCP SDK, a VS Code extension, or a wrapper around an existing client.
Acceptance Criteria
References