Skip to content

Interview Prep

Clients

CpBruceMeena/interview-prep

MCP Client Implementations¶

clients/
├── python_client.py      # Command-line client for the servers in ../servers
├── claude_config.json    # Example desktop-client (mcpServers) configuration
└── __init__.py

Python client (`python_client.py`)¶

Built on the SDK v2 mcp.Client, which replaces v1's stdio_client + ClientSession + initialize() boilerplate. Against a server on the 2026-07-28 spec it skips the handshake (it probes server/discover, then every request carries its version and capabilities in _meta); against older servers it falls back to initialize automatically.

cd ai-engineering/mcp
python -m clients.python_client --server calculator --list        # tools, resources, prompts
python -m clients.python_client --server calculator --add 5 3
python -m clients.python_client --server rag --query "What is RAG?"
python -m clients.python_client --server calculator --interactive # /tools /call /read ...

In your own code:

from mcp import Client, StdioServerParameters

async with Client(StdioServerParameters(command="python", args=["-m", "servers.calculator_server"])) as client:
    tools = await client.list_tools()
    result = await client.call_tool("add", {"a": 10, "b": 20})
    print(result.content[0].text, result.structured_content)   # "30.0" {'result': 30.0}

async with Client("https://mcp.example.com/mcp") as client:    # Streamable HTTP
    ...

result.content is what you would show a model; result.structured_content is what program code should parse. Check result.is_error before using either.

Desktop configuration (`claude_config.json`)¶

Hosts such as Claude Desktop and IDE agents read an mcpServers map: each entry is a command the host launches as a stdio subprocess. The host does not set your working directory, so tell Python where the package is:

{
  "mcpServers": {
    "calculator": {
      "command": "/absolute/path/to/venv/bin/python",
      "args": ["-m", "servers.calculator_server"],
      "env": { "PYTHONPATH": "/absolute/path/to/interview-prep/ai-engineering/mcp" }
    }
  }
}

Remote servers are added by URL instead (Streamable HTTP, usually with OAuth); the exact config key varies by host, so check its docs.