mcp-tool-dev

MCP Tool Creation

Create individual MCP tool functions that follow Anthropic's tool design best practices and the Model Context Protocol specification.

What MCP Tools Are

An MCP tool is a single function exposed via the Model Context Protocol that an LLM can invoke. Each tool has:

• Name: Verb-noun format (search_documents, get_user, create_issue)
• Description: 3-4 sentences explaining what the tool does, when to use it, and when not to
• Input schema: JSON Schema defining parameters with clear descriptions
• Handler: Async function that processes the input and returns MCP content blocks

Tools can be standalone (single-file utilities) or part of an MCP server (grouped related tools).

When to Use This Skill

• Creating a single tool function for an existing or new MCP server
• Designing tool schemas and descriptions
• Implementing tool input validation and error handling
• Adding capabilities to a FastMCP server

When NOT to Use

• Building a complete multi-tool MCP server from scratch — use mcp-server-dev instead
• Creating Claude Code agents, skills, or commands — use their respective creation skills

Tool Anatomy

Every MCP tool consists of four parts:

1. Name

Use snake_case with verb-noun pattern. Be specific — search_files_by_content beats search.

2. Description

Write 3-4 sentences covering:

1. What the tool does (capability statement)
2. When to use it (primary use cases)
3. When NOT to use it (scope boundaries, suggest alternatives)
4. Key behavior notes (pagination, rate limits, return format)

Good descriptions prevent misuse and reduce wasted calls. Include parameter semantics — if "query" means regex vs full-text vs exact match, say so.

3. Input Schema

Define parameters using JSON Schema with:

• Clear description for each parameter explaining expected format and semantics
• enum constraints where values are known
• default values for optional parameters
• Required vs optional distinction