Model Context Protocol (MCP)

Modelcontextprotocol.io

• docs
• Building effective agents
• python-sdk source

Reference Implementations

FastMCP

• source
• docs

A high-level, fast, Pythonic way to build MCP servers and clients

# install
pip install fastmcp

# verify version
fastmcp version

FastMCP version:                                                            2.5.2
MCP version:                                                                1.9.2
Python version:                                                            3.13.2
Platform:                            Linux-6.8.0-60-generic-x86_64-with-glibc2.35
FastMCP root path:              ~/anaconda3/envs/mcp/lib/python3.13/site-packages

# dev install
git clone https://github.com/jlowin/fastmcp.git
cd fastmcp
uv sync

Fast Agent

• source
• story

A framework with complete, end-to-end tested MCP Feature support including Sampling, which helps create and interact with sophisticated Agents and Workflows in minutes. Both Anthropic (Haiku, Sonnet, Opus) and OpenAI models (gpt-4o/gpt-4.1 family, o1/o3 family) are supported

Demos

FastMCP - Gemini

A fastmcp tutorial using Gemini.

• MCP server:
  • mcp_server.py: collection of tools/resources like calculator, trigonometry function, yahoo finance API call
• MCP clients:
  • mcp_client_simple.py: simple client with rule-based query parsing
  • mcp_client_llm.py: client with LLM-based query parsing
  • mcp_client_llm_resource.py: client with LLM parsing and calling tool/resource (see readme_tool_resource.md for details)

see readme_gemini.md for more details

Setup

conda create -n mcp
conda activate mcp
git clone https://github.com/digital-duck/mcp_demo.git

cd mcp_demo
pip install -r requirements.txt

# in 1st terminal
python mcp_server.py

# in 2nd terminal
# python mcp_client_simple.py
python mcp_client_llm.py
# python mcp_client_llm_resource.py

Streamlit MCP app

streamlit run st_mcp_app.py

FastMCP - AWS

see readme_aws.md for more details