Component Implementation: Resources

Our application implements resources as readable documents rather than callable functions - which is the standard way to define them in the MCP protocol.

A client reads a resource to get background context — things like dataset notes, metric definitions, or static statistics — that help an LLM interpret tool results correctly. Our server exposes two resources from server/components/resources.py.

Imports and Registration Shell

The structure mirrors tools.py exactly: a single register_resources function receives mcp and uses it to decorate two inner functions. Keeping the same pattern across all component files makes the codebase easy to navigate.

# components/resources.py
import json

from fastmcp import FastMCP


def register_resources(mcp: FastMCP):
    ...

Resource 1: `netflix://guide` — Dataset Notes

The resource URI netflix://guide is a custom scheme that identifies this document uniquely within the server.

    @mcp.resource("netflix://guide")
    def get_data_guide() -> str:
        """Netflix data guide with important notes and usage information."""
        guide_data = {
            "about": "Netflix viewing data for movies from June 2021 to December 2024",
            "important_notes": [
                "All tools return WEEKLY viewing data only (7-day periods)",
                "Weekly rankings show top 10 positions only (rank 1-10)",
                "Use 'hours_viewed' as primary metric - it's more reliable than 'views'",
                "Movie title searches are partial match and case-insensitive",
                "The 'views' metric can be NULL for some periods",
                "Cumulative weeks in top 10 tracks how many total weeks a title ranked",
            ],
            "data_characteristics": {
                "content_type": "Movies only",
                "time_range": "June 2021 - December 2024",
                "granularity": "Weekly snapshots (7-day periods)",
                "geographic_scope": "Global Netflix viewing data",
            },
            "available_metrics": {
                "hours_viewed": "Total viewing hours in the period (most reliable metric)",
                "views": "Number of views (may be NULL for some periods)",
                "view_rank": "Position in weekly top 10 (1-10, or NULL if outside top 10)",
                "cumulative_weeks_in_top10":