Ollama-MCP-Streamable-http-Server by prncher - MCP Server

🧠 MCP Streamable HTTP Client & Server with Ollama and Selenium

This project demonstrates how to build a Model Context Protocol (MCP) system using:

🧰 MCP Server for tool execution
🌐 Streamable HTTP MCP Client (Python)
🦙 Ollama running qwen2.5-coder:7b for reasoning
🖥️ Selenium for browser automation

The architecture enables natural language instructions like:

“Go to MDN and print HTML DOM API page content.”

The MCP Client uses Ollama to generate a structured plan of actions and parameters.
The MCP Server executes the plan in a live Selenium browser and streams results back in real time.

🧰 Getting Started

1. Install Ollama

Download and install Ollama 0.12.3 from the official site:
👉 https://ollama.com/download/windows

2. Pull the Model

After installing Ollama, pull the required model:

ollama pull qwen3-coder:30b

💡 You can also use qwen2.5-coder:7b for a faster, lightweight setup.

3. Install Python Dependencies

Install uv (a fast Python package installer):

pip install uv

4. Create and Activate a Virtual Environment

🪟 Windows

uv venv
.venv\Scripts\activate

🍎 macOS / 🐧 Linux

uv venv
source .venv/bin/activate

5. Install the Project in Editable Mode

This installs dependencies and links the local project for development:

uv pip install -e .

📝 VS Code Tip: You may need to manually select the .venv Python interpreter to resolve imports.

6. Run the MCP Server

Start the MCP server to handle tool execution:

uv run src\server.py

7. Run the MCP Client

In a separate terminal, run the MCP Client with a natural language instruction:

uv run src\client.py "Goto MDN and print HTML DOM API page content"

The client will:

Send the prompt to Ollama
Receive the action plan
Forward it to the MCP server
Stream the results back in real time

🧭 Key Features

First-of-its-kind Streamable HTTP MCP Client in Python
Real-time streaming of execution logs and results
Clean separation of reasoning (LLM) and execution (MCP Server)
Extensible toolbox of browser actions (launch, click, scroll, extract, etc.)

🏗️ Architecture

User → MCP Client → Ollama → MCP Client → MCP Server → Selenium
                         ↑                   ↓
                       Streamed responses ←───

📌 See the full article in for a deep dive.

📝 Tools Implemented

Launch Browser
Click Element / Selector
Type Text
Scroll Page
Get Page Content
Get DOM Structure
Take Screenshot
Extract Data
Close Browser

🚀 Example

Prompt: "Go to MDN and print HTML DOM API page content"

The client generates a plan via Ollama, sends it to the server, and streams back the full DOM contents in real time.

📂 Repositories

🧪 Future Work

Testing with larger models (qwen2.5-coder:30b) to improve selector accuracy
Adding persistent event storage for replay and durability
Expanding the toolbox with more advanced browser actions

prncher/Ollama-MCP-Streamable-http-Server