Chef‑Agent Knowledge‑Graph Cooking Assistant

A streaming AI “Chef” agent that uses LangGraph workflows, MCP tools, and a Neo4j‑backed recipe knowledge graph to answer cooking queries, update or ingest recipes, and remember user preferences.

Agent Graph Architecture Image

🚀 Features

• Interactive streaming conversation via FastMCP + FastAPI

• Graph‑driven recipe storage & updates (Neo4j + langchain_neo4j + LLMGraphTransformer)

• Tool support for:

  • web_search (Tavily/DuckDuckGo)

  • web_scraper (FireCrawl + BeautifulSoup fallback)

  • execute_python sandboxed code

  • graph_query (natural‑language → Cypher)

  • ingest_url_to_graph (scrape & ingest new recipes)

• Memory via in‑process store (or Redis) to personalize sessions

• Auto‑summarization of long chats with a short‑term summarizer

Related MCP server: HowToCook-MCP Server

📦 Prerequisites

• Python 3.10+

• Neo4j 4.4+ (standalone or Docker)

• Redis Stack (if using RedisStore/checkpointer)

Environment Variables

Create a .env file at project root and set all of the following:

# Multi‑provider LLM keys
GOOGLE_API_KEY=
GROQ_API_KEY=
CEREBRAS_API_KEY=

# Search & scraping
TAVILY_API_KEY=
E2B_API_KEY=
FIRECRAWL_API_KEY=

# Langfuse observability
LANGFUSE_PUBLIC_KEY=
LANGFUSE_SECRET_KEY=
LANGFUSE_HOST=

# Neo4j connection
NEO4J_URI=
NEO4J_USERNAME=
NEO4J_PASSWORD=
NEO4J_DATABASE=

# Redis (optional)
DB_URI=redis://localhost:6379/0

🔧 Installation

1. Clone repo

   git clone https://github.com/your-org/chef-agent.git
   cd chef-agent

2. Create & activate a virtual env

   python -m venv .venv
   source .venv/bin/activate
   pip install -r requirements.txt

3. Set your .env as above.

⚙️ Running the MCP Server

python mcp_server.py

• Exposes MCP tools at http://127.0.0.1:8000/mcp

• Health check: GET /health

⚙️ Running the Agent

python agent.py

• Connects to MCP server

• Builds a LangGraph StateGraph workflow:

  1. assistant: generates initial answer, sets has_final once [FinalAnswer]: appears

  2. tools: invokes any needed tools (web_search, graph_query, etc.)

  3. update_graph → graph_update_tool_calling: decides & applies graph updates

  4. finalize_answer: produces final user‑facing recipe plan

  5. write_memory → summarization_node: saves memory & summarizes

• Streaming output: prints incremental responses

📂 Code Structure

.
├── agent.py            # Main agent orchestration & graph workflow
├── mcp_server.py       # FastAPI + FastMCP tool definitions
├── graphDB.py          # GraphDB wrapper (Neo4j + LLMGraphTransformer)
├── schemas.py          # Pydantic models: Recipe, Profile, UpdateGraphDecision
├── scrapper.py         # Web scraper & Markdown converter
├── prompts/
│   ├── SYSTEM_PROMPT.txt
│   ├── decision_prompt.txt
│   ├── decision_prompt_2.txt
│   ├── conversation_prompt.txt
│   └── summarization_prompt.txt
├── requirements.txt
├── .env
└── README.md

🛠️ Customization

• Switch LLM: in agent.py change provider="google" to "groq" or another supported model.

• Enable Redis for persistence: swap InMemoryStore/Saver with AsyncRedisStore/Saver and set DB_URI.

• Extend tools: add new @mcp.tool() functions in mcp_server.py.

🐞 Troubleshooting

• Graph connectivity: confirm Neo4j credentials & network reachability.

Future Work

• Multi-agent: multiple agents can be run in parallel & share memory.

• Distributed: multiple instances of the agent can be run on different machines.

• Multilingual: Support for multiple languages.

• Multimodal: Support for video/image based analysis for clear instructions.

• Multi-modal: Support for voice based analysis for clear instructions.

• Security: Add authentication & authorization.

• Voice: Support for voice based analysis for clear instructions.

Contributing

Contributions are welcome! To contribute:

1. Fork the repository.

2. Create a new branch.

3. Submit a pull request with your changes.

Contact

For any questions or suggestions, feel free to contact on below Contact details:

• Om Nagvekar Portfolio Website, Email: https://omnagvekar.github.io/ , omnagvekar29@gmail.com

• GitHub Profile:

  • Om Nagvekar: https://github.com/OmNagvekar

📜 License

This project is licensed under the GPL-3.0 license.