MCP-style LangChain + Ollama Python project that:

Runs completely free and locally

Accepts MCP-style { action, params } via a Flask API

Routes to a local LLM (via Ollama) for answers

File:

app.py ← Flask API (MCP-style server)

│

├── app.py ← Flask API (MCP-style server)-> Flask web server

├── mcp_agent.py ← LangChain logic: interprets actions -> LangChain agent

├── requirements.txt ← Python dependencies

└── README.md ← Project overview

MCP-based LangChain server

The LangChain agent received the correct action

It executed it using your logic (from mcp_agent.py)

It responded with the correct time back to Postman

Next Steps: Test with other action values

Add new functions to mcp_agent.py

Or integrate this with a frontend

adding more intelligent tasks or connecting it with a UI!

Llama models

Explore https://ollama.com/library

Run gemma:2b

ollama run gemma:2b

gemma:2b uses just ~2–3 GB RAM.

Run mistral

ollama run mistral

mistral only needs ~4 GB of RAM and still performs quite well for general tasks.

Run llama3

ollama run llama3

Ollama runs models like llama3 entirely on your machine, in memory.

Even though it's optimized, LLaMA 3 still needs at least ~6 GB of RAM free, and ideally more (8–12 GB total system RAM recommended).

List downloaded models:

ollama list

Pull other models:

ollama pull mistral

ollama pull codellama

Stop running model:

ollama stop

Start your MCP server:

python app.py

You need to install the latest LangChain with community modules, specifically:

pip install langchain-community

Recommended Full Setup (for safety):

If you're using LangChain with Ollama, it's best to ensure these are installed:

pip install langchain langchain-community langchain-core langchainhub

pip install ollama

gemma:2b ,llama3 has cut off date

Static Knowledge: These models are trained on data available up to a certain point (e.g., mid-2023 for many models).

They don’t know anything that happened after their training cut-off date.

Want Real News in a LangChain/Ollama Setup?

You can combine the local model with:

RAG (Retrieval-Augmented Generation), where you fetch live news via API (like NewsAPI or Google News),

Then pass that info as context to the LLM to summarize or answer based on it.

LangChain + News API integration using local Ollama model.