resource-design_skill

Overview

This skill helps you design MCP resources for xmcp so LLMs can consume content reliably. It guides you through an interactive info-gathering workflow, then generates a properly routed resource file with metadata, schema (if needed), and a handler.

How this skill works

I first prompt you for key inputs: resource name, type (static, dynamic, or file-based), URI pattern, content details, and whether widget rendering is needed. Based on your answers I create a TypeScript file under src/resources/ using Next.js-style file routing, include optional metadata and zod schema for dynamic parameters, and implement a handler that returns the appropriate content type. I also run through a checklist to ensure routing, metadata, and return types match xmcp expectations.

When to use it

• Creating a simple static resource to expose docs or configuration to LLMs.
• Building a dynamic resource that accepts parameters (e.g., user profiles or reports).
• Wrapping file-based content such as logs or binary assets for LLM access.
• Conforming file layout to xmcp routing conventions for predictable URIs.
• Adding OpenAI widget metadata or other vendor extensions for richer rendering.

Best practices

• Use kebab-case for resource filenames and clear, descriptive names in metadata.
• Prefer explicit mimeType and size in metadata when serving binary or non-text content.
• For dynamic resources, define a zod schema and call .describe() on parameters for clear documentation.
• Place route groups in parentheses to exclude them from the URI and use [param] for dynamic segments.
• Return { text: string } for clear text responses, and use Uint8Array with mimeType for binary blobs.

Example use cases

• Static config: src/resources/(config)/app.ts → config://app with plain text config.
• User profile: src/resources/(users)/[userId]/profile.ts → users://[userId]/profile returning JSON.
• Monthly report: src/resources/reports/[year]/[month]/summary.ts → reports://[year]/[month]/summary generating dynamic summaries.
• File-based logs: src/resources/logs/[fileName].ts reading a log file and returning its contents with correct mimeType.
• Widget-enabled docs: add OpenAI metadata to enable rich rendering of a documentation resource.