Defining tools with MCP

Overview

• Building an MCP (Model Context Protocol) server to add functionality to an existing CLI chatbot
• Server will provide tools that Claude can use to interact with documents
• Implementation goes in mcpserver. py file in the project root

MCP Python SDK Benefits

• Official MCP Python SDK simplifies server and tool creation
• Creates MCP server with just one line of code
• Automatically generates JSON schemas for tools—no manual schema writing needed
• Significantly reduces boilerplate compared to manual tool definition

Tool Definition Pattern

• Use @MCP. tool decorator to define a tool
• Provide tool name and description
• Define function with typed parameters using Pydantic Field
• SDK automatically converts function signature into proper JSON schema

Tool 1: Read Document

• Name: read_doc_contents
• Purpose: Retrieve contents of a document by ID
• Parameter: doc_id (string) - ID of document to read
• Implementation: Look up document in docs dictionary and return its contents
• Error handling: Raise ValueError if document ID doesn't exist

Tool 2: Edit Document

• Name: edit_document
• Purpose: Modify document content using find-and-replace
• Parameters:
• doc_id (string) - ID of document to edit
• old_string (string) - Text to find (must match exactly, including whitespace)
• new_string (string) - Replacement text
• Implementation: Use string . replace() method on document contents
• Error handling: Raise ValueError if document ID doesn't exist

Key Imports

• Import Field from pydantic to define tool parameters with descriptions

Takeaways

• MCP Python SDK simplifies server creation to one line of code and automatically generates JSON schemas for tools, eliminating manual boilerplate
• Tools are defined using the @MCP. tool decorator with typed Pydantic Field parameters that the SDK converts to proper JSON schemas
• Two core tools needed: read_doc_contents (retrieves document by ID) and edit_document (modifies content via find-and-replace)
• Both tools require error handling to raise ValueError when a document ID doesn't exist
• Import Field from pydantic to add descriptions to tool parameters