A LangChain + Ollama agent that talks to a FastMCP server using Llama 3.1 8B

This project demonstrates a fully functional AI agent that:

• Connects to a FastMCP JSON-RPC server over HTTP
• Discovers tools via tools.list
• Uses Llama 3.1 8B (via Ollama) to decide when to call tools
• Calls tools like lunch_menu(date) and report_absence(date, reason)
• Returns natural language responses

No empty args. No missing parameters. 100% working.

mcp-school-tools/
├── mcp_server_fixed.py       # MCP JSON-RPC server (manual registry)
├── mcp_agent_final_fixed.py  # LangChain + Ollama agent (structured JSON)
├── README.md                 # This file
└── requirements.txt          # Python dependencies

1. Python 3.11+
2. Ollama installed and running
3. Llama 3.1 8B model pulled

ollama pull llama3.1:8b

git clone https://github.com/yourusername/mcp-school-tools.git
cd mcp-school-tools

pip install -r requirements.txt

requirements.txt:

fastapi
uvicorn
langchain
langchain-ollama
langchain-core
pydantic
requests

python mcp_server_fixed.py

Server runs at: http://127.0.0.1:8000

python mcp_agent_final_fixed.py

### Ollama Start Up

Ollama serve

You: what is for lunch on 2025-11-09
LLM response: {'tool': 'lunch_menu', 'args': {'date': '2025-11-09'}}
Sending to MCP: lunch_menu(args={'date': '2025-11-09'})

Agent: Veggie Burger, Sweet Potato Fries, Orange Wedges

You: report absence on 2025-11-12 because I'm sick
LLM response: {'tool': 'report_absence', 'args': {'date': '2025-11-12', 'reason': "I'm sick"}}

Agent: Absence reported for 2025-11-12. Reason: I'm sick

1. Agent sends initialize → notifications/initialized → tools.list
2. Server returns full JSON schema for each tool
3. Agent prompts Llama 3.1 with escaped JSON examples
4. LLM outputs structured JSON: {"tool": "...", "args": {...}}
5. Agent validates args with Pydantic
6. Agent calls MCP server with tools.call
7. Server executes tool and returns result

• Check MCP server logs for incoming tools.call
• Check agent logs for Sending to MCP
• Use curl to test server directly:

curl -X POST http://127.0.0.1:8000 \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","method":"tools.call","params":{"tool":"lunch_menu","args":{"date":"2025-11-08"}},"id":1}'

1. Add function in mcp_server_fixed.py
2. Register in TOOLS dict with parameters
3. Add Pydantic model in mcp_agent_final_fixed.py
4. Add to TOOL_SCHEMAS

• Web UI (Streamlit/Gradio)
• Conversation memory
• Async support
• Authentication
• More tools (grades, schedule, etc.)

MIT

Built with love for hackathons, demos, and learning.

Let me know when you're ready to deploy!