MCP Dev Setup: Anaconda, uv, and Claude

The Model Context Protocol (MCP) lets AI models talk to external tools, files, and APIs through one open standard. Before we build MCP servers and clients, we need a dependable local toolchain. That means a Python distribution, a local LLM runtime, a JavaScript runtime for the many npm-based MCP servers, the Claude Desktop host, and a fast package manager.

In this blog, we install Anaconda, Ollama, Node.js, Claude Desktop, and uv. We default to Windows commands and label the Linux and macOS equivalents clearly. At the end, we build a reproducible uv project that loads API keys from a .env file.

Installing Anaconda

Anaconda is a Python and R data-science distribution with built-in package and environment management. It gives us conda, a clean way to isolate the Python version this course uses (3.11+).

Download the installer for your operating system from the official Anaconda download page.

System requirements: Windows 10+ (64-bit), macOS 10.13+, or 64-bit Linux, each needing about 5 GB of free space.

Windows installation

1. Run the downloaded .exe file (right-click → Run as administrator if prompted).
2. Follow the installation wizard.
3. When offered, check Add Anaconda to PATH so conda works from any terminal.
4. Complete the installation (10-15 minutes).

On macOS: open the downloaded .pkg file and follow the installer prompts.

On Linux: run the installer script and accept the license terms.

bash ~/Downloads/Anaconda3-2024.02-1-Linux-x86_64.sh

Verify and create an environment

conda --version
conda list

Create and activate a dedicated environment for this course:

conda create -n ml python=3.11
conda activate ml

Installing Ollama

Ollama runs large language models locally on our machine. So we can build private MCP agents without sending data to a cloud API. This course uses local models such as qwen3, gpt-oss, and the nomic-embed-text embedding model.

Windows installation

Download the installer from the official Ollama site and run the .exe. On Windows and macOS, Ollama runs as a background service automatically after install.

On macOS: download from ollama.com, or use Homebrew:

brew install ollama

On Linux:

curl -fsSL https://ollama.com/install.sh | sh

Pull and run models

ollama pull llama3.2:3b
ollama pull qwen3
ollama pull nomic-embed-text

Run a model interactively and list what you have installed:

ollama run llama3.2:3b
ollama list

Verify the local API is responding (Ollama serves on port 11434):

ollama --version
curl http://localhost:11434/api/generate -d "{\"model\": \"llama3.2:3b\", \"prompt\": \"Why is the sky blue?\"}"

On Linux/macOS: the same curl works with single quotes around the JSON body:

curl http://localhost:11434/api/generate -d '{"model": "llama3.2:3b", "prompt": "Why is the sky blue?"}'

Installing Node.js

Many MCP servers ship as npm packages and run with npx (for example the filesystem, Playwright, and Airbnb servers used later in this course). Node.js provides that JavaScript runtime.

Windows installation

Download the LTS installer from nodejs.org and run it, or use a Windows package manager:

winget install OpenJS.NodeJS

On macOS:

brew install node

On Linux (Ubuntu/Debian):

sudo apt update
sudo apt install nodejs npm

Verify

node --version
npm --version
node -e "console.log('Hello Node.js!')"

Installing Claude Desktop

Claude Desktop is Anthropic's desktop app. It also doubles as an MCP host. That means it can connect to multiple MCP servers at once and call their tools directly inside a conversation. We use it heavily in the data-analysis and Claude Desktop sections of this course.

Download it from the official Claude download page.

Windows installation

1. Download Claude-Setup.exe.
2. Run the installer and follow the wizard.
3. Launch Claude from the Start Menu, sign in with your Anthropic account, and accept the terms.

On macOS: open the downloaded .dmg, drag Claude into your Applications folder, and launch it.

uv vs pip: a faster package manager

uv is a modern, very fast Python package installer and project manager written in Rust. It is a drop-in replacement for pip and pip-tools. It is the tool this course uses to manage dependencies.

Why uv over pip:

• 10-100× faster installs and dependency resolution.
• Built-in virtual environment management, with no separate venv step.
• Lockfiles by default (uv.lock) for reproducible builds.
• Smarter dependency resolution for complex conflicts.
• Consistent behavior across Windows, macOS, and Linux.

Installing uv on Windows

powershell -c "irm https://astral.sh/uv/install.ps1 | iex"

On Linux/macOS:

curl -LsSf https://astral.sh/uv/install.sh | sh

You can also install it with pip into your active environment:

pip install uv

pip workflow vs uv workflow

The old pip workflow needs us to create and activate a virtual environment by hand:

python -m venv myenv
myenv\Scripts\activate
pip install requests numpy
pip freeze > requirements.txt

On Linux/macOS: activate with source myenv/bin/activate instead of myenv\Scripts\activate.

The uv workflow creates the project, environment, and lockfile for us:

uv init myproject
cd myproject
uv add requests numpy
uv sync

uv add records the dependency in pyproject.toml and updates uv.lock automatically. uv sync installs exactly what the lockfile pins.

Creating the course project with uv

Initialize a project for the course and add the libraries we need across every section:

uv init mcp-mastery-claude-and-langchain
cd mcp-mastery-claude-and-langchain

This scaffolds a project:

mcp-mastery-claude-and-langchain/
├── pyproject.toml    # Project configuration
├── uv.lock           # Lockfile (auto-generated)
├── main.py
└── README.md

Add the full dependency set used throughout the course:

uv add fastmcp "mcp[cli]" mcp-use langchain langchain-community langchain-ollama langchain-openai langchain-chroma langchain-mcp-adapters langgraph chromadb faiss-cpu streamlit fastapi pandas numpy scipy matplotlib pypdf2 requests python-dotenv

Run a Python script inside the project environment without manually activating anything:

uv run python main.py

Loading API keys from a .env file

Some sections use cloud services (the Claude API, OpenAI embeddings for the deployed server, and optional LangSmith tracing). Store secrets in a .env file at the project root and load them with python-dotenv. Never hard-code keys in source.

Create a .env file:

OPENAI_API_KEY=your-openai-key-here
CLAUDE_API_KEY=your-anthropic-key-here
LANGSMITH_API_KEY=your-langsmith-key-here

A minimal main.py that loads and verifies the keys:

import os
from dotenv import load_dotenv
load_dotenv()

def main():
    print("OPENAI_API_KEY:", os.getenv("OPENAI_API_KEY"))
    print("CLAUDE_API_KEY:", os.getenv("CLAUDE_API_KEY"))
    print("LANGSMITH_API_KEY:", os.getenv("LANGSMITH_API_KEY"))
    print("Hello from mcp-mastery-claude-and-langchain!")

if __name__ == "__main__":
    main()

Run it:

uv run python main.py

OPENAI_API_KEY: your-openai-key-here
CLAUDE_API_KEY: your-anthropic-key-here
LANGSMITH_API_KEY: your-langsmith-key-here
Hello from mcp-mastery-claude-and-langchain!

Setup checklist

We now have the full toolchain installed and verified:

• Anaconda: conda --version works; an ml environment is active.
• Ollama: ollama list shows qwen3 / nomic-embed-text; the API answers on port 11434.
• Node.js: node --version and npm --version print versions.
• Claude Desktop: installed and signed in (Windows/macOS).
• uv: a project with pyproject.toml + uv.lock runs uv run python main.py.

With the environment ready, we can move on. Next, we will understand what MCP actually is and how its client, host, and server pieces fit together in Introduction to Model Context Protocol.