Canada's Food Guide MCP Server

# IMPLEMENTATIONS.md Future improvements To-Do list, tasks, Project wishlists, Notes ## August 12, 2025 ### -- Rate Limiting & Web Scraping Improvements [] Implement exponential backoff with jitter for CNF web scraping - Current 0.5s fixed delay insufficient for bulk operations - Add randomized delays (2-4 seconds) to mimic human behavior - Implement exponential backoff on HTTP 429 responses [] Add request chunking for bulk_get_cnf_macronutrients operations - Break large bulk requests (20 foods) into smaller chunks (3-5 foods) - Add progressive delays between chunks - Reduce concurrent workers from 3 to 1-2 for better server respect [] Enhanced error handling and retry mechanisms - Distinguish between rate limits, network errors, and data errors - Implement circuit breaker pattern for repeated failures - Better caching strategy for failed requests [] Configuration improvements for production stability - Increase CNF_RATE_LIMIT from 0.5s to 2-3s - Add environment variables for rate limit customization - Add monitoring for rate limit hit frequency [] Long-term solutions to reduce web scraping dependency - Investigate Health Canada official API access - Build offline CNF database cache for common foods - Implement MCP resources for frequently requested nutrients ## June 14, 2025 [] CNF database updated robot.txt limits web api calls to prevent bloat - Now only 2 of, say 4, recipes are being fetched. Alternatives --> Add mcp.resource or mcp.tool for table values for common foods --> Possibly add offline database for Canada's nutrient file --> Update rate for api calls. ## June 10, 2025 ### TO-DO [] Update the outdate CLI to include db, EER, and CNF compatibility. LLM's waste water, and this in itself can have a standalone value too. ### -- Improvements [] Macros Analysis is the most resource consuming tool set, and it also tends to time out on the Nutrient File's side. An API key here would really efficient. [] Combine these tools (potentially), to make tool calling efficient, and reduce tool calls: - [] calculate_recipe_nutrition_summary and query_recipe_macros_table. The first tool simple parses and sets up the table for query recipes - [] store_recipe_in_temp_tables and simple_recipe_setup. the setup function parses the stored ingredients, they should be ideally combined, so to reduce LLM tool calls. [] refactor and breakdown cnf_tools.py file. The current size is too big, and contains helper functions [] Add mermaid diagram to implementations.md to show dataflow. [] Simplify installaion with an npm or smithery installation packaging ### -- Potential ideas [] Extend DRI support to include micro nutrients, vitamins and minerals too [] add tools to allow users to store temp_recipe_ info to permanent database for ingredients they keep in stock [] add functionality to vary how long you like to store recipe information ## Notes -> Current LLM clients already have great web crawlers, though the current mcp idea is a good example of using html webscraping, it might be an outdated feature in light of current crawlers and their developments. Maybe a good justification of using a smaller model on a phone in a kitchen. -> Need to think of tests, and evaluating tool call frequencies to make this server more sustainable.