MCP Server Development

Scope

TRIGGER: The model must activate when building MCP (Model Context Protocol) servers

SPECIALIZATION: FastMCP framework (Python, decorator-based) FALLBACK: Generic Python SDK, TypeScript SDK also covered

COVERAGE:

• Generic MCP protocol (all implementations)
• Agent-centric design principles
• FastMCP framework (Python, decorator-based, Pydantic validation)
• TypeScript/Node MCP SDK (Zod validation)
• Evaluation creation for testing server quality
• Production deployment and packaging
• Security, performance, and observability patterns

EXCLUSIONS:

• Low-level MCP transport layer details (handled by SDKs/frameworks)
• Client-side MCP implementations

High-Level Workflow

TRIGGER: The model must follow this 4-phase workflow when building MCP servers

Phase 1: Deep Research and Planning

1.1 Understand Agent-Centric Design Principles

RULE: The model must design tools for AI agents, not just API wrappers

PRINCIPLES:

Build for Workflows, Not Just API Endpoints:

• Don't simply wrap existing API endpoints
• Build thoughtful, high-impact workflow tools
• Consolidate related operations (e.g., schedule_event that both checks availability and creates event)
• Focus on tools that enable complete tasks, not just individual API calls
• Consider what workflows agents actually need to accomplish

Optimize for Limited Context:

• Agents have constrained context windows - make every token count
• Return high-signal information, not exhaustive data dumps
• Provide "concise" vs "detailed" response format options
• Default to human-readable identifiers over technical codes (names over IDs)
• Consider the agent's context budget as a scarce resource

Design Actionable Error Messages:

• Error messages should guide agents toward correct usage patterns
• Suggest specific next steps: "Try using filter='active_only' to reduce results"
• Make errors educational, not just diagnostic
• Help agents learn proper tool usage through clear feedback

Follow Natural Task Subdivisions:

• Tool names should reflect how humans think about tasks
• Group related tools with consistent prefixes for discoverability
• Design tools around natural workflows, not just API structure

Use Evaluation-Driven Development:

• Create realistic evaluation scenarios early
• Let agent feedback drive tool improvements
• Prototype quickly and iterate based on actual agent performance

1.2 Study MCP Protocol Documentation

RESOURCE: https://modelcontextprotocol.io/llms-full.txt PURPOSE: Complete MCP specification and guidelines

RESEARCH TOOL PREFERENCE (in order):

1. Preferred: mcp__Ref__ref_search_documentation(query="Model Context Protocol specification") - High-fidelity verbatim documentation
2. Alternative: mcp__exa__get_code_context_exa(query="MCP protocol server implementation examples") - Code context and examples
3. Fallback: WebFetch - Use only when MCP tools don't provide needed content

RATIONALE: MCP tools provide higher fidelity (verbatim source) compared to WebFetch (AI summaries)

Web resource access, definitive guide for getting accurate data for high quality results

1.3 Study Framework Documentation

DECISION_TREE:

IF implementing in Python THEN
  - Load FastMCP patterns from references/development-guidelines.md
  - Load Python SDK documentation from official source
  - Focus on decorator-based patterns and Pydantic validation

ELSE IF implementing in TypeScript/Node THEN
  - Load TypeScript patterns from references/typescript-mcp-server.md
  - Load TypeScript SDK documentation from official source
  - Focus on registerTool patterns and Zod validation

ELSE
  - Load generic best practices from references/mcp-best-practices.md
  - Adapt to target language/framework

REFERENCES_IN_SKILL:

• MCP Best Practices - Universal MCP guidelines
• FastMCP Development Guidelines - Python FastMCP specialization
• TypeScript MCP Server Guide - TypeScript/Node implementation
• FastMCP Community Practices - Mid-2025+ patterns
• Prompts and Templates - Prompt system configuration
• Example Projects - Real-world implementations
• Evaluation Guide - Testing server quality

1.4 Exhaustively Study API Documentation

RULE: The model must read through ALL available API documentation for target service

REQUIREMENTS:

• Official API reference documentation
• Authentication and authorization requirements
• Rate limiting and pagination patterns
• Error responses and status codes
• Available endpoints and their parameters
• Data models and schemas

RESEARCH TOOL HIERARCHY (use in this order):

1. MCP Ref tool: mcp__Ref__ref_search_documentation for official API documentation (verbatim, high-fidelity)
2. MCP exa tool: mcp__exa__get_code_context_exa for code examples and SDK usage patterns
3. MCP exa web search: mcp__exa__web_search_exa for general research with LLM-optimized results
4. Fallback: WebFetch only when MCP tools don't work

RATIONALE: MCP tools provide superior accuracy, precision, and fidelity for technical documentation

1.5 Create Comprehensive Implementation Plan

PLAN_COMPONENTS:

Tool Selection:

• List the most valuable endpoints/operations to implement
• Prioritize tools that enable the most common and important use cases
• Consider which tools work together to enable complex workflows

Shared Utilities and Helpers:

• Identify common API request patterns
• Plan pagination helpers
• Design filtering and formatting utilities
• Plan error handling strategies

Input/Output Design:

• Define input validation models (Pydantic for Python, Zod for TypeScript)
• Design consistent response formats (JSON and Markdown)
• Design configurable levels of detail (Detailed or Concise)
• Plan for large-scale usage (thousands of users/resources)
• Implement character limits and truncation strategies (e.g., 25,000 tokens)

Error Handling Strategy:

• Plan graceful failure modes
• Design clear, actionable, LLM-friendly, natural language error messages
• Consider rate limiting and timeout scenarios
• Handle authentication and authorization errors

Phase 2: Implementation

RULE: The model must implement following language-specific best practices

For Python (FastMCP)

Python Project Setup:

RULE: The model must activate the python3-development skill before setting up Python MCP server projects

CONSTRAINT: The python3-development skill contains:

• Python project layouts (src/ vs flat)
• pyproject.toml structure with uv
• Modern Python 3.11+ patterns
• Package structure best practices
• Build/publishing guides
• Lessons learned for successful Python project design

ACTIVATION:

Skill(command: "python3-development")

The model must defer to python3-development for general Python project structure.

Basic Server Structure:

from fastmcp import FastMCP
from pydantic import Field
from typing import Annotated

mcp = FastMCP("service_mcp")

@mcp.tool()
def tool_name(
    param: Annotated[str, Field(description="Parameter description")]
) -> dict:
    """Tool description for the AI."""
    return {"result": "value"}

if __name__ == "__main__":
    mcp.run()  # Defaults to STDIO transport

Key Patterns:

• Use @mcp.tool() decorator for tools
• Use Annotated[type, Field(...)] for parameter validation
• First line of docstring becomes tool description
• Return dict for structured output
• Use async def for I/O-bound operations

See: FastMCP Development Guidelines for complete Python patterns

For TypeScript/Node

Basic Server Structure:

import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
import { z } from "zod";

const server = new McpServer({
  name: "service-mcp-server",
  version: "1.0.0",
});

const InputSchema = z
  .object({
    param: z.string().describe("Parameter description"),
  })
  .strict();

server.registerTool(
  "tool_name",
  {
    title: "Tool Title",
    description: "Tool description for the AI",
    inputSchema: InputSchema,
    annotations: { readOnlyHint: true },
  },
  async (params) => {
    return { content: [{ type: "text", text: "result" }] };
  },
);

const transport = new StdioServerTransport();
await server.connect(transport);

Key Patterns:

• Use registerTool with complete configuration
• Use Zod schemas with .strict()
• Explicitly provide title, description, inputSchema, annotations
• Type all parameters and return values

See: TypeScript MCP Server Guide for complete TypeScript patterns

Phase 3: Review and Refine

3.1 Code Quality Review

CHECKLIST:

• DRY Principle: No duplicated code between tools
• Composability: Shared logic extracted into functions
• Consistency: Similar operations return similar formats
• Error Handling: All external calls have error handling
• Type Safety: Full type coverage (Python type hints, TypeScript types)
• Documentation: Every tool has comprehensive docstrings/descriptions

3.2 Test and Build

Important: MCP servers are long-running processes. Running them directly causes your process to hang indefinitely.

Safe Testing:

• Use evaluation harness (see Phase 4)
• Run server in tmux to keep it outside main process
• Use timeout when testing: timeout 5s python server.py

For Python:

python -m py_compile your_server.py  # Verify syntax

For TypeScript:

npm run build  # Must complete without errors

Phase 4: Create Evaluations

RULE: The model must create comprehensive evaluations to test server effectiveness

RESOURCE: Evaluation Guide

Purpose: Evaluations test whether LLMs can effectively use your MCP server to answer realistic, complex questions.

Process:

1. Tool Inspection - List available tools and understand capabilities
2. Content Exploration - Use READ-ONLY operations to explore available data
3. Question Generation - Create 10 complex, realistic questions
4. Answer Verification - Solve each question yourself to verify answers

Requirements: Each question must be:

• Independent (not dependent on other questions)
• Read-only (only non-destructive operations required)
• Complex (requiring multiple tool calls and deep exploration)
• Realistic (based on real use cases humans would care about)
• Verifiable (single, clear answer that can be verified by string comparison)
• Stable (answer won't change over time)

Output Format:

<evaluation>
  <qa_pair>
    <question>Complex question requiring multiple tool calls</question>
    <answer>Single verifiable answer</answer>
  </qa_pair>
</evaluation>

Evaluation Scripts:

Located in ./scripts/:

• evaluation.py - Evaluation harness
• connections.py - MCP connection utilities
• requirements.txt - Python dependencies
• example_evaluation.xml - Example evaluation

Usage:

pip install -r scripts/requirements.txt
export ANTHROPIC_API_KEY=your_api_key

python scripts/evaluation.py \
  -t stdio \
  -c python \
  -a my_mcp_server.py \
  evaluation.xml

Quick Reference

FastMCP (Python) Quick Start

from fastmcp import FastMCP
from pydantic import Field
from typing import Annotated

mcp = FastMCP("my-server")

@mcp.tool()
def search_items(
    query: str,
    limit: Annotated[int, Field(ge=1, le=100)] = 10
) -> dict:
    """Search for items matching the query."""
    results = perform_search(query, limit)
    return {"results": results, "count": len(results)}

@mcp.resource("data://config")
def get_config() -> dict:
    return {"setting": "value"}

@mcp.prompt()
def explain_topic(topic: str) -> str:
    return f"Explain the concept of '{topic}'"

if __name__ == "__main__":
    mcp.run()  # STDIO transport
    # OR: mcp.run(transport="http", port=8000)

TypeScript/Node Quick Start

import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
import { z } from "zod";

const server = new McpServer({
  name: "my-server",
  version: "1.0.0",
});

const SearchSchema = z
  .object({
    query: z.string(),
    limit: z.number().int().min(1).max(100).default(10),
  })
  .strict();

server.registerTool(
  "search_items",
  {
    title: "Search Items",
    description: "Search for items matching the query",
    inputSchema: SearchSchema,
    annotations: { readOnlyHint: true },
  },
  async (params) => {
    const results = await performSearch(params.query, params.limit);
    return {
      content: [
        {
          type: "text",
          text: JSON.stringify({ results, count: results.length }, null, 2),
        },
      ],
    };
  },
);

const transport = new StdioServerTransport();
await server.connect(transport);

Best Practices Summary

MANDATORY_PRACTICES:

Tool Design:

• Design workflow-oriented tools, not API endpoint wrappers
• Use descriptive names with service prefix: {service}_{action}_{resource}
• Optimize for AI context window efficiency
• Provide actionable error messages

Input/Output:

• Support both JSON and Markdown response formats
• Implement pagination for list operations
• Enforce CHARACTER_LIMIT (typically 25,000) with truncation
• Use human-readable identifiers where appropriate

Validation:

• Python: Use Pydantic Field() with constraints
• TypeScript: Use Zod schemas with .strict()
• Validate all inputs against schema
• Sanitize file paths and external identifiers

Error Handling:

• Don't expose internal errors to clients
• Provide clear, actionable error messages
• Use ToolError (Python) for business logic errors
• Handle timeouts and rate limits gracefully

Security:

• Validate file paths against allowed directories
• Use confirmation flags for destructive operations
• Set destructiveHint annotation for state-changing tools
• Rate limit expensive operations
• Store secrets in environment variables

Performance:

• Use async for I/O-bound operations
• Cache repeated queries using lru_cache or similar
• Stream large responses in HTTP mode
• Extract common functionality into reusable functions

Deployment:

• Package as .mcpb for Claude Desktop distribution
• Provide manifest.json with user_config fields
• Support environment variable configuration
• Test with evaluation harness before release

Resources

COMPREHENSIVE_REFERENCES:

mcp-best-practices.md

Universal MCP guidelines including naming conventions, response formats, pagination, security, and compliance requirements. Use for all MCP implementations regardless of language.

development-guidelines.md

Complete FastMCP development guide covering decorators, Pydantic validation, async patterns, error handling, Context parameters, annotations, transport options, and production deployment.

typescript-mcp-server.md

Complete TypeScript/Node implementation guide covering project structure, registerTool patterns, Zod validation, error handling, and production build process.

community-practices.md

Mid-2025+ best practices including .mcpb packaging, security by design patterns, observability and testing approaches, performance tuning, caching, ecosystem compatibility, and agent orchestration patterns.

prompts-and-templates.md

FastMCP prompt and template system covering @mcp.prompt decorator, system instructions for tool use, configuration for AI-native tools, and prompt engineering for MCP servers.

example-projects.md

Real-world FastMCP implementations demonstrating best practices and patterns from Ultimate MCP Server (AI Agent OS), Hugging Face MCP server, browser automation servers, data/DevOps integrations, coding assistants, and templates/aggregators.

evaluation-guide.md

Complete guide for creating comprehensive evaluations to test whether LLMs can effectively use your MCP server to answer realistic, complex questions. Includes question guidelines, answer requirements, evaluation process, output format, examples, and verification process.

Standalone Operation

CONSTRAINT: This skill is completely standalone with no external dependencies

VERIFICATION:

• ✅ All generic MCP best practices included
• ✅ All FastMCP Python patterns included
• ✅ All TypeScript/Node patterns included
• ✅ All evaluation creation guidance included
• ✅ All security, performance, and observability patterns included
• ✅ All community practices and .mcpb packaging included
• ✅ All scripts and evaluation harness included
• ✅ All reference files self-contained
• ✅ No references to external skills or resources

USAGE: Load this skill to access complete, comprehensive MCP server development guidance without needing any other skills or external resources.