Scraps Kitchen MCP Server

Your AI assistant's cooking brain. Scraps Kitchen gives any AI agent persistent, household-aware kitchen intelligence — pantry tracking, dietary safety, recipe management, shopping lists, and more.

Unlike generic chatbot memory, Scraps maintains structured cooking data: what's in your fridge, who you cook for, their allergies and preferences, your recipe history with cook notes, and your kitchen equipment. Any MCP-compatible agent can tap into this to have genuinely personalized cooking conversations.

Live server: https://api.scraps.kitchen/mcp (Streamable HTTP) Web app: scraps.kitchen

Quick Start

Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "scraps-kitchen": {
      "type": "streamable-http",
      "url": "https://api.scraps.kitchen/mcp"
    }
  }
}

VS Code / Cursor

Add to your MCP settings:

{
  "servers": {
    "scraps-kitchen": {
      "type": "streamable-http",
      "url": "https://api.scraps.kitchen/mcp"
    }
  }
}

On first use, you'll be prompted to authenticate via OAuth. You can create an account directly in the OAuth flow — a one-time code is sent to your email for verification. No need to visit scraps.kitchen separately.

Related MCP server: yes_chef_mcp

Tools (27)

Meal Suggestions

Pantry (6 tools)

Shopping List (5 tools)

Household (4 tools)

Equipment (4 tools)

Recipes (6 tools)

Authentication

All tools require authentication. Scraps uses OAuth 2.1 with PKCE — most MCP clients handle this automatically. You'll see a login prompt on first connection, where you can sign in or create a new account with email OTP verification.

Scopes:

Discovery: https://api.scraps.kitchen/.well-known/mcp.json

How Agents Use Scraps

Scraps is a data layer, not a reasoning engine. It stores structured kitchen intelligence that makes any AI assistant better at cooking conversations:

1. Read the kitchen context — get_pantry, get_household, get_equipment, get_recipes

2. Use your own LLM to reason about what to cook, considering allergies, preferences, and available ingredients

3. Write back results — mark_ingredients_used, add_recipe_note, rate_recipe_for_diner

Every interaction makes the data richer. More cook history, more diner feedback, more pantry accuracy — which makes the next conversation better.

Example Workflows

"What should I cook tonight?"

1. get_pantry → see available ingredients

2. get_household → check dietary constraints and allergens

3. get_equipment → know what tools are available

4. get_recipes(status: "cooked") → see what's worked before

5. Agent reasons about a meal suggestion using all this context

After cooking:

1. mark_ingredients_used(["chicken thighs", "ginger", "soy sauce"]) → update pantry

2. add_recipe_note(recipe_id, "Added extra garlic, seared 2 min longer") → build cook history

3. rate_recipe_for_diner(recipe_id, diner_id, { loved_it: true }) → track preferences

Grocery planning:

1. get_pantry(stale_only: true) → items past freshness window

2. get_shopping_list → existing list

3. add_shopping_list_item("eggs") → add what's needed

4. After shopping: add_pantry_items([...]) → restock, clear_checked_shopping_items → clean up list

Links

• Web app: scraps.kitchen

• MCP endpoint: https://api.scraps.kitchen/mcp

• Discovery: https://api.scraps.kitchen/.well-known/mcp.json

• Official MCP Registry: registry.modelcontextprotocol.io

License

MIT