🧮 Simple MCP Server (built with FastMCP)

A "Hello World" MCP (Model Context Protocol) server built with FastMCP — a small calculator server you can talk to over in-memory, HTTP, and STDIO transports, plus a LangGraph ReAct agent that uses its tools.

It exposes:

Type	Name	Description
🛠️ Tool	add(a, b)	Add two integers
🛠️ Tool	subtract(a, b)	Subtract b from a
📄 Resource	file:///endpoint/{name}	Returns a template message
📄 Resource	file://endpoint2/{name}	Reads a real file from ./path/
💬 Prompt	review_code(code)	A reusable "review this code" template

---

Project structure

simple-MCP-server-built-with-FastMCP/
├── server.py            # the FastMCP server (tools + resources + prompt)
├── client_inmemory.py   # test in the same process (no network)
├── client_http.py       # connect over HTTP
├── client_stdio.py      # spawn the server as a subprocess (STDIO)
├── agent_example.py     # LangGraph ReAct agent that uses the tools
├── path/                # files served by the file resource
│   ├── README.txt
│   └── examples.txt
├── requirements.txt
├── .env.example         # API key template (for the agent only)
└── README.md

---

Setup

cd simple-MCP-server-built-with-FastMCP

python -m venv .venv
# Windows (PowerShell):
.\.venv\Scripts\Activate.ps1
# macOS / Linux:
source .venv/bin/activate

pip install -r requirements.txt

---

Run it

1. Quick test — in-memory (no network, no API key)

python client_inmemory.py

Calls the tools, lists them, reads both resources, and renders the prompt.

2. HTTP transport

Terminal A (start the server):

python server.py http        # serves http://127.0.0.1:8000/mcp

Terminal B (run the client):

python client_http.py

3. STDIO transport

python client_stdio.py       # spawns server.py itself — nothing to start first

4. ReAct agent (needs an API key)

# one-time: copy the template and add a key
Copy-Item .env.example .env   # then edit .env  (GROQ_API_KEY is free)
python agent_example.py

The agent loads the MCP tools, then answers "What is 8 + 7? Use the tools." by actually calling the add tool.

💡 Groq is free — get a key at console.groq.com. The agent uses Groq if GROQ_API_KEY is set, otherwise OpenAI.

---

How MCP transports differ

• In-memory — client and server share one Python process (simplest; for testing).
• HTTP — server runs as a web service; clients connect by URL (good for remote/shared).
• STDIO — client launches the server as a child process and talks over stdin/stdout (good for local tools).

The same tools/resources/prompt work over all three — only the "wire" changes.

---

Credits

Based on the IBM Skills Network lab "Hello World of MCP Servers." Built with FastMCP, LangGraph, and langchain-mcp-adapters.