DEV Community

Akshay Gore
Akshay Gore

Posted on

Bare Minimum Local MCP Setup

#ai #mcp #agents #programming

This post is about creating a simple local setup to understand MCP hands-on

The Flow

The LLM is the brain, the agent is the worker, MCP is a standard adapter, and the queried system is the actual tool being plugged in — but the brain never talks to MCP directly, only through the worker.

Breaking it down:

  1. LLM = the brain that reasons and decides what needs to happen
  2. Agent = the worker that carries out actions using that brain's decisions
  3. MCP server = a universal adapter/socket that lets the worker plug into any tool without custom wiring for each one
  4. The queried thing (database, calendar, GitHub, etc.) = the actual appliance behind that socket

GitHub Link

Don't talk, show

Lets build the entire system on local system

1. Run the model locally 

ollama start
ollama run llama3.2:1b
ollama show llama3.2:1b
Enter fullscreen mode Exit fullscreen mode 
  Model
    architecture        llama
    parameters          1.2B
    context length      131072
    embedding length    2048
    quantization        Q8_0

  Capabilities
    completion
    tools

  License
    LLAMA 3.2 COMMUNITY LICENSE AGREEMENT
    Llama 3.2 Version Release Date: September 25, 2024
Enter fullscreen mode Exit fullscreen mode 
curl http://localhost:11434/api/ps
{"models":[{"name":"llama3.2:1b","model":"llama3.2:1b","size":1510389841,"digest":"baf6a787fdffd633537aa2eb51cfd54cb93ff08e28040095462bb63daf552878","details":{"parent_model":"","format":"gguf","family":"llama","families":["llama"],"parameter_size":"1.2B","quantization_level":"Q8_0"},"expires_at":"2026-07-30T13:26:33.201017+05:30","size_vram":1510389841,"context_length":4096}]}
Enter fullscreen mode Exit fullscreen mode

2. Setting up an agent

2.1 Creating a python env 

pyenv local 3.12.0
Enter fullscreen mode Exit fullscreen mode 
python3 --version
Python 3.12.0
Enter fullscreen mode Exit fullscreen mode

2.2 Creating python venv for installing agent 

python -m venv venv
source venv/bin/activate
Enter fullscreen mode Exit fullscreen mode

3. Running a MCP server with help of docker which helps to expose our simple notes. This is the location where MCP comes to know about our stuff 

docker build -t mcp-notes-server .
docker run -d \
  --name mcp-notes \
  -p 8765:8765 \
  -v $(pwd)/notes.txt:/data/notes.txt \
  mcp-notes-server
Enter fullscreen mode Exit fullscreen mode 
CONTAINER ID   IMAGE                                 COMMAND                  CREATED          STATUS          PORTS                                                                                                                                  NAMES
9d5c24e619ad   mcp-notes-server                      "python mcp_server.py"   28 seconds ago   Up 27 seconds   0.0.0.0:8765->8765/tcp, [::]:8765->8765/tcp                                                                                            mcp-notes
Enter fullscreen mode Exit fullscreen mode

4 Running a basic python agent - (Important) 

python agent.py
Enter fullscreen mode Exit fullscreen mode

4.1 Agent is up and running waiting for input 

Agent with MCP (Docker) tool access. Type 'exit' to quit.

[MCP] Connected to container. Available tools: ['read_notes']

You:
Enter fullscreen mode Exit fullscreen mode

The Verdict

Here we prompt the model asking about our notes using the tool exposed by the MCP server

Agent with MCP (Docker) tool access. Type 'exit' to quit.

[MCP] Connected to container. Available tools: ['read_notes']

You: Can you check whats in my notes using the tools read_notes
Agent:
[AGENT] Sending request to Ollama at 22:47:54.174
TOOL_CALL: read_notes
[MCP] Model requested tool call: read_notes
[MCP] Tool result:
Meeting with Sarah at 3pm.
Buy groceries: eggs, milk, bread.
Finish the agent project this week.
Added few more comments to test notes

Agent:
[AGENT] Sending request to Ollama at 22:47:54.360
You need help organizing your tasks and reminders in your local notes.txt file?
You:
Enter fullscreen mode Exit fullscreen mode

Key Point
The python agent created above talks to ollama which is hosted locally. The ollama becomes aware about mcp because the config is saved in agent.py
Key Point - OLLAMA_URL = "http://localhost:11434/api/chat" (agents.py)

GitHub Link

Top comments (0)