Defining tools with MCP

MCP Server Development Study Notes

**I. Core Concepts**

• Purpose of MCP Server: To add external functionality (tools) to the existing CLI chatbot, allowing it to interact with specific data and perform actions.
• Implementation Location: The server logic is placed in MCP server. py within the root project directory.
• MCP Python SDK: The official SDK is used to simplify server setup and tool definition.
• Benefit: The SDK automatically generates the necessary JSON schema for tools, eliminating the need for manual schema writing.
• In-Memory Data Storage: Documents are stored in a dictionary (docs) where the key is the document ID and the value is the document's content.

**II. Tool Definition Process**

• Tool Definition: Tools are defined by specifying a name, a description, and required arguments.
• Tool Implementation: The actual logic is written in a Python function decorated with @mcp. tool.
• Required Components:
• Tool Name (e. g. , read_doc_contents).
• Tool Description (explains when and how the tool should be used).
• Function Signature (defines arguments and their types).

**III. Tool Implementations**

**A. Read Document Contents Tool**

• Goal: Retrieve the content of a specified document.
• Function: read_document(doc_id)
• Logic:
• Check if doc_id exists as a key in the docs dictionary.
• If not found, raise a ValueError (e. g. , "doc with ID doc_id not found").
• If found, return the corresponding document content (docs[doc_id]).

**B. Edit Document Tool**

• Goal: Perform a simple find-and-replace operation within a document.
• Function: edit_document(doc_id, old_string, new_string)
• Logic:
• Check if doc_id exists in the docs dictionary.
• If not found, raise a ValueError.
• If found, update the document content: docs[doc_id] = docs[doc_id]. replace(old_string, new_string).

Takeaways

• The MCP Server adds external functionality (tools) to a CLI chatbot, allowing it to interact with specific data.
• The MCP Python SDK simplifies server setup and automatically generates the necessary JSON schema for tools.
• Documents are stored in memory using a dictionary (docs) where the key is the document ID and the value is the content.
• Tools are implemented as Python functions decorated with @mcp. tool, requiring a name, description, and function signature.
• Tools like read_document and edit_document directly access the docs dictionary to retrieve or modify document content.