Defining tools with MCP

MCP Server Implementation Study Notes

**I. Project Goal and Setup**

• Objective: To add an MCP (Model Context Protocol) server to a CLI chatbot to provide document management capabilities.
• Location: The server implementation resides in the MCP server. py file within the root project directory.
• Initial State: The server is set up with an in-memory collection of documents stored in a dictionary (docs), where keys are document IDs and values are the document contents.

**II. Key Technology: MCP Python SDK**

• Function: The official MCP Python SDK simplifies the creation of the MCP server (often requiring only one line of code).
• Tool Definition: The SDK makes defining tools easy by allowing developers to write the tool's logic directly in Python.
• Process: When a tool is defined using the SDK, it automatically generates the necessary tool JSON schema, which can then be passed to Bedrock.

**III. Tool Implementation Concepts**

• Tool Definition Structure: A tool is defined by specifying its name, a description, and its required arguments (including their types and descriptions).
• Tool 1: Read Document Contents (read_doc_contents)
• Purpose: Retrieves the content of a specified document.
• Input: doc_id (string) – The ID of the document to read.
• Logic: Looks up the doc_id in the in-memory docs dictionary and returns the corresponding value.
• Error Handling: Must check if doc_id exists in docs; if not, raise a ValueError stating the document was not found.
• Tool 2: Edit Document (edit_document)
• Purpose: Modifies the content of an existing document by performing a string replacement.
• Inputs:
• doc_id (string) – The ID of the document to be edited.
• old_string (string) – The exact text to be replaced.
• new_string (string) – The new text to insert.
• Logic: Finds the document using doc_id and executes a find-and-replace operation on its content.
• Error Handling: Must check if doc_id exists in docs; if not, raise a ValueError.

**IV. Summary of SDK Benefits**

• The MCP Python SDK significantly reduces the complexity of defining tools compared to manually writing out large JSON schemas.
• It handles the conversion of Python function definitions (including decorators and Pydantic fields) into the required JSON schema format for the LLM (Claude/Bedrock).

Takeaways

• The MCP server adds document management capabilities to the CLI chatbot.
• The MCP Python SDK simplifies tool creation by automatically generating JSON schemas from Python function definitions.
• Tools are defined by specifying a name, description, and required arguments (e. g. , doc_id, old_string).
• Key tools include read_doc_contents (retrieves content) and edit_document (performs string replacement).
• The server manages documents using an in-memory dictionary (docs) and requires robust error handling for missing document IDs.