MCP File System Agent

A local agentic file-system assistant built with LangChain, Ollama, FastMCP, and MCP.

The project demonstrates how an LLM can autonomously select and use tools exposed by an MCP server to perform file-related operations such as reading, listing, writing, and searching files.

Architecture

                         User
                           │
                           ▼
                  ┌─────────────────┐
                  │  LangChain      │
                  │     Agent       │
                  └────────┬────────┘
                           │
                           ▼
                  ┌─────────────────┐
                  │   Ollama LLM    │
                  │   Qwen3 1.7B    │
                  └────────┬────────┘
                           │
                    Tool selection
                           │
                           ▼
                ┌──────────────────────┐
                │ LangChain MCP Adapter│
                └──────────┬───────────┘
                           │
                       MCP / HTTP
                           │
                           ▼
                ┌──────────────────────┐
                │     FastMCP Server   │
                │   localhost:8000/mcp │
                └──────────┬───────────┘
                           │
             ┌─────────────┼─────────────┐
             ▼             ▼             ▼
        read_file     list_files    write_file
             │
             ▼
       Local File System

Features

• MCP-based tool architecture
• Local FastMCP server using HTTP transport
• LangChain agent with tool calling
• Local LLM inference through Ollama
• PDF file reading
• DOCX file reading and writing
• File listing with extension filtering
• Searching inside PDF and DOCX files
• Conversation state using LangGraph checkpointing
• Automatic tool selection by the LLM

Tech Stack

• Python
• LangChain
• LangGraph
• Ollama
• Qwen3 1.7B
• FastMCP
• Model Context Protocol (MCP)
• langchain-mcp-adapters
• PyPDF
• python-docx

Project Structure

MCP/
│
├── tests/
│   └── sample.docx
│
├── fs_mcp.py
├── main.py
├── requirements.txt
└── README.md

Installation

1. Clone the repository

git clone https://github.com/jibixn/File-System-Tools-MCP
cd MCP

2. Install Python dependencies

pip install -r requirements.txt

3. Install Ollama

Install Ollama from the official website and make sure it is running locally.

Then pull the model:

ollama pull qwen3:1.7b

You can verify that the model is available with:

ollama list

Running the Project

The project uses two processes because the MCP server communicates with the client over HTTP.

Terminal 1 — Start the MCP server

Run this command from the project root:

python fs_mcp.py

The server should be available at:

http://127.0.0.1:8000/mcp

Terminal 2 — Start the agent

Open another terminal in the project root:

python main.py

You can then enter requests such as:

Read the file in tests folder named sample.docx and provide me the summary.

or:

List all PDF files in the tests folder.

or:

Write "Hello, World!" to tests/output.docx.

Example Agent Flow

For a request such as:

Read the file in tests folder named sample.docx and provide me the summary.

the agent performs the following:

User request
     │
     ▼
LLM analyzes request
     │
     ▼
LLM selects read_file
     │
     ▼
LangChain MCP Adapter
     │
     ▼
MCP HTTP request
     │
     ▼
FastMCP read_file()
     │
     ▼
python-docx reads file
     │
     ▼
Tool result returned to agent
     │
     ▼
LLM summarizes content
     │
     ▼
Final response

Requirements

Python 3.11+ is recommended.

Ollama must be installed and running locally.

The Qwen model must be available:

ollama pull qwen3:1.7b

You can also use a model through an inference provider.

Dependencies

The project's direct dependencies are:

fastmcp==3.4.7
langchain==1.3.14
langchain-mcp-adapters==0.3.2
langchain-ollama==1.1.0
langgraph-checkpoint==4.2.0
pypdf==6.14.2
python-docx==1.2.0

Future Improvements

• Add support for more file formats
• Add file deletion and directory creation tools
• Add stronger path validation and sandboxing
• Add authentication for remote MCP servers
• Add streaming responses
• Add richer document parsing
• Add persistent conversation storage
• Add additional MCP servers for databases, GitHub, or web search
• Improve tool-selection reliability with larger local models

Learning Goals

This project demonstrates the interaction between:

LLM
 ↓
LangChain Agent
 ↓
Tool Calling
 ↓
MCP Client
 ↓
MCP Protocol
 ↓
FastMCP Server
 ↓
Python Functions

It is intended as a practical example of building an agentic application with Model Context Protocol (MCP) and locally hosted LLMs.