MCP Client

If you're building a custom agent app, you implement or use an MCP client to consume any compliant server without custom integration code.

An MCP client is a universal remote control that pairs with any MCP-compatible device after a one-time setup.

Scenario

Your custom agent app's MCP client connects to three servers at startup — filesystem, GitHub, and Postgres — and merges their tool lists for the LLM.

Solution

In Model Context Protocol, apply MCP Client to this scenario: Your custom agent app's MCP client connects to three servers at startup — filesystem, GitHub, and Postgres — and merges their tool lists for the LLM. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Do this before moving to the next module — reading alone is not enough.

Open the Code Walkthrough below and run it locally. Change one parameter related to MCP Client (e.g. model, temperature, top_k, or tool name), observe the difference in output, and write 2–3 sentences explaining what changed.

Highlighted lines show where MCP Client happens in the code.

1from mcp import ClientSession, StdioServerParameters  # import dependencies2from mcp.client.stdio import stdio_client  # import dependencies3
4server_params = StdioServerParameters(command="npx", args=["-y", "@modelcontextprotocol/server-filesystem", "/workspace"])5async with stdio_client(server_params) as (read, write):  # key line for MCP Client6    async with ClientSession(read, write) as session:  # key line for MCP Client7        await session.initialize()8        tools = await session.list_tools()9        result = await session.call_tool("read_file", {"path": "README.md"})

• Treating MCP Client as a black box without evaluation
• Ignoring cost and latency in production
• Skipping error handling for mcp client

Quick recap — the most important points from this module.