traceverse-community/food-delivery-evals

Food Delivery — SmolTrace Tasks

Part of the TraceVerse Community open evaluation & observability stack.

A SmolTrace-format evaluation suite of 111 tasks for agents calling food-delivery MCP servers. Pairs with the food-delivery-mcp Space — every expected_tool here is implemented there.

---

Dataset info

• Tasks: 111
• Format: SmolTrace evaluation format
• Tools covered: search_restaurants, view_menu, place_order, track_delivery, apply_promo, rate_restaurant, contact_driver
• Companion MCP server: traceverse-community/food-delivery-mcp
• License: Apache 2.0

Usage with SmolTrace

from datasets import load_dataset

dataset = load_dataset("traceverse-community/food-delivery-evals")
# smoltrace-eval --model openai/gpt-4o-mini --dataset-name traceverse-community/food-delivery-evals

Schema

Column	Type	Description
id	string	food-delivery_<tool>[_<suffix>]_batch<n>_<m>
prompt	string	Task description
expected_tool	string	One of: search_restaurants, view_menu, place_order, track_delivery, apply_promo, rate_restaurant, contact_driver
difficulty	string	easy / medium / hard
agent_type	string	tool (ToolCallingAgent) or code (CodeAgent)
expected_keywords	list[string]	2-7 keywords expected in plan/output

Example rows

# Easy / tool agent
{
    "id": "food-delivery_search_restaurants_batch0_0",
    "prompt": "I'm craving Italian food. Can you find some Italian restaurants near me?",
    "expected_tool": "search_restaurants",
    "difficulty": "easy",
    "agent_type": "tool",
    "expected_keywords": ["Italian", "restaurants", "search"]
}

# Hard / code agent
{
    "id": "food-delivery_apply_promo_multiple_conditions_batch16_416",
    "prompt": "I want to order from 'Global Eats'. Check if any of these promos are applicable: 'SAVE15' for orders over $50, or 'FREESHIP' for orders above $30. Apply the one with the biggest discount, or the free shipping if both are applicable.",
    "expected_tool": "apply_promo",
    "difficulty": "hard",
    "agent_type": "code",
    "expected_keywords": ["Global Eats", "SAVE15", "FREESHIP", "discount", "free shipping", "conditional logic"]
}

Distribution

• By tool: search_restaurants 26 · place_order 20 · view_menu 19 · track_delivery 13 · apply_promo 13 · rate_restaurant 10 · contact_driver 10
• By difficulty: easy 75 · medium 19 · hard 17
• By agent type: tool 85 · code 26

---

Prompt template

Save the YAML below as prompt_template.yaml and pass via system_prompt_path to smolagents.ToolCallingAgent or CodeAgent.

system_prompt: |-
  You are an expert food-delivery assistant who can solve any task using tool calls. You have MCP tools to search restaurants, view menus, place orders, track deliveries, apply promo codes, rate restaurants, and contact drivers.

  Each step is Action (tool call) → Observation. Finish with `final_answer`.

  Example:

  Task: "Order a Margherita pizza from a top-rated Italian place."
  Action:
  {"name": "search_restaurants", "arguments": {"cuisine": "Italian", "min_rating": 4.0}}
  Observation: "Found 3 Italian restaurants: 'Pizza Palace' (4.5 stars), ..."
  Action:
  {"name": "view_menu", "arguments": {"restaurant_name": "Pizza Palace"}}
  Observation: "Menu: Margherita Pizza ($12), ..."
  Action:
  {"name": "place_order", "arguments": {"restaurant_name": "Pizza Palace", "items": "Margherita Pizza"}}
  Observation: "Order placed. Order ID: #12345. ETA: 30 minutes."
  Action:
  {"name": "final_answer", "arguments": {"answer": "Pizza ordered from Pizza Palace. Order ID #12345, ETA 30 min."}}

  Available tools:
  {%- for tool in tools.values() %}
  - {{ tool.to_tool_calling_prompt() }}
  {%- endfor %}

  Rules: always tool-call; pass values not variables; never repeat with identical args; finish with final_answer.

planning:
  initial_plan: |-
    Survey known facts, list facts to look up, then write a step-by-step plan ending with <end_plan>.
    Task: {{task}}

system_prompt_code: |-
  You are an expert food-delivery assistant solving tasks with Python code. Use Thought / Code / Observation cycles, finish with `final_answer(...)`.

  Example:

  Task: "Find a 4-star+ Italian restaurant and order a pizza."
  Thought: Search Italian, min_rating 4.0; pick highest; view menu; order.
  {{code_block_opening_tag}}
  results = search_restaurants(cuisine="Italian", min_rating=4.0)
  print(results)
  {{code_block_closing_tag}}
  Observation: {"results": [{"id": "r-001", "name": "Pizza Palace", "rating": 4.5}, ...]}
  Thought: Order Margherita from Pizza Palace.
  {{code_block_opening_tag}}
  menu = view_menu(restaurant_id="r-001")
  order = place_order(restaurant_id="r-001", items="m-101")
  final_answer(f"Order placed: {order['order_id']}, ETA {order['eta_minutes']} min.")
  {{code_block_closing_tag}}

  Tools (callable as Python functions):
  {{code_block_opening_tag}}
  {%- for tool in tools.values() %}
  {{ tool.to_code_prompt() }}
  {% endfor %}
  {{code_block_closing_tag}}

  Rules: Thought + Code each step; values not variables; state persists across blocks; allowed imports: {{authorized_imports}}.

For the full unabridged template (with planning / managed_agent / final_answer sections), see MCP-1st-Birthday/smoltrace-food-delivery-tasks — it's the same SmolTrace template family.

---

Reproducing this dataset

hf clone datasets/traceverse-community/food-delivery-evals
cd food-delivery-evals
python3 generate.py > tasks.jsonl

---

Part of the TraceVerse Community stack

Project	Where
Reference agents + docs	food-delivery-agents
MCP server (this dataset's companion)	food-delivery-mcp
Trace viewer	TraceMind
Instrumentation	genai-otel-instrument
Evaluation framework	SmolTrace

Sibling datasets:

• instamart-evals — grocery / quick-commerce tasks
• dineout-evals — dine-in / reservation tasks

Citation

@misc{traceverse-food-delivery-evals,
  author    = {TraceVerse Community},
  title     = {Food Delivery Evals: a SmolTrace evaluation dataset for food-delivery MCP agents},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/traceverse-community/food-delivery-evals}
}

Acknowledgement

The SmolTrace evaluation format and prompt-template structure originated in the work of MCP-1st-Birthday on the TraceMind MCP Server's synthetic data generation. We extend that work with TraceVerse-Community-curated tasks aligned to the food-delivery-mcp tool surface.