paprika-mcp-server: Instructions file for Claude Code

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Project Overview

Paprika MCP Server is a Model Context Protocol (MCP) server that provides access to Paprika Recipe Manager data. It enables AI assistants to search and retrieve recipes stored in Paprika format.

The project has two main functions:

1. MCP Server: Runs as an MCP server exposing recipe search/list/get tools via stdio transport
2. Recipe Unpacker: Converts .paprikarecipes archive files into individual JSON files

Commands

Development

# Install dependencies
pnpm install

# Build the project (bundles into single dist/index.cjs file)
pnpm build

# Build minified production bundle (~56% smaller)
pnpm build:prod

# Run directly in development (without building)
pnpm start

# Lint
pnpm lint

# Lint and auto-fix
pnpm fix

Using the CLI

The CLI has two main commands:

# Start MCP server
./dist/index.js mcp --recipes-dir <path> [--server-name <name>] [--verbose]

# Unpack .paprikarecipes archive to individual JSON files
./dist/index.js unpack --input <file> --output <dir> [--verbose]

Environment variables:

• PAPRIKA_RECIPES_DIR: Default recipes directory (defaults to .recipes in CWD)
• PAPRIKA_SERVER_NAME: Default server name (defaults to "paprika")

Architecture

Core Components

RecipeStore (src/RecipeStore.ts)

• In-memory recipe database using RxDB with memory storage
• Loads recipes from a directory of JSON files on initialization
• Provides list(), search(query, fields?), and getCount() methods
• Search is client-side filtering (case-insensitive substring matching)
• Uses RxDB schema validation with Ajv

CLI System (src/cli/)

• Built on @rushstack/ts-command-line for robust CLI parsing
• PaprikaCommandLine: Main CLI parser that registers actions
• Actions:
  • McpAction: Starts MCP server, loads recipes, registers tools
  • UnpackAction: Extracts .paprikarecipes (zip) containing gzipped JSON files