Phase 8Module 6 of 25

Tools

Function calling was reinvented in every framework. MCP tools are the same idea, with one discovery story for every client.

A tool is a labeled button on a machine. The label (description) is what the operator reads before pressing it.

Visual Workflows

Start here — scroll inside each diagram frame to explore, then use + / − to zoom up to 200% if needed.

Overview

100%

Loading diagram...

Scroll inside the frame to explore · use + / − to zoom up to 200%

100%

Loading diagram...

Scroll inside the frame to explore · use + / − to zoom up to 200%

GitHub example

100%

Loading diagram...

Scroll inside the frame to explore · use + / − to zoom up to 200%

search_code is a tool. The repo README would be a resource.

Key Takeaways

1.MCP tools are actions with a name, description, and JSON Schema. The LLM picks a tool from the description, not from your Python name.
2.call_tool sends arguments; the server returns CallToolResult content. One MCP tool works in Cursor, Claude Desktop, and your custom agent.
3.Treat descriptions as UI copy for the model. If two tools sound the same, the model will flip a coin.
4.Namespace tools when you attach several servers (github/search_code).

Learn elsewhere

→Resources vs tools — previous module
→Routing tools in an agent — Integrate MCP with Agent

Real Example

Scenario

A GitHub MCP server exposes create_issue, search_code, and get_pull_request. Cursor's agent calls search_code, then opens an issue with create_issue.

What you would do

In Model Context Protocol, apply Tools to this scenario: A GitHub MCP server exposes create_issue, search_code, and get_pull_request. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Commands

Commands to Remember

Tool = name + description + inputSchema
call_tool(name, arguments)
Description is the LLM's instruction manual
Namespace tools when many servers overlap

Cheat Sheet

Quick recap

quick ref

•Tools are verbs / actions
•JSON Schema for arguments
•Honest descriptions prevent wrong calls
•Return errors as results, keep the process up

Common Mistakes

✕Two tools with almost the same description
✕Optional arguments that are actually required
✕A mega-tool that does five unrelated jobs