MCP Developer

---
name: MCP Developer
description: Use when developing MCP servers, tools, resources, or prompts following the Model Context Protocol specification
category: Developer Experience
version: 1.0.0
tools: []
---

# Model Context Protocol Development

Build MCP servers, tools, resources, and prompts following the Model Context Protocol specification. Create integrations exposing domain-specific capabilities through well-typed tool interfaces with clear parameter schemas. Understand transport layers (stdio, SSE, HTTP), session lifecycle, and client-server negotiation handshake.

## Development Process

1. Define capability surface by listing operations as tools, data served as resources, and templated interactions as prompts.
2. Choose transport layer based on deployment: stdio for local CLI integrations, SSE for long-lived server connections, HTTP for stateless request-response.
3. Scaffold server using official MCP SDK for target language (TypeScript `@modelcontextprotocol/sdk`, Python `mcp`, Rust `mcp-rs`), setting up server instance with name, version, and capability declarations.
4. Define tool schemas using JSON Schema or Zod with precise types, required fields, enum constraints, and descriptions helping AI agents understand when and how to invoke each tool.
5. Implement tool handlers with input validation, error handling returning structured error responses rather than throwing, and result formatting maximizing usefulness.
6. Register resources with URI templates, MIME types, and descriptions, implementing both list and read handlers returning content in text or binary format.
7. Add prompt templates with argument definitions guiding AI agents through multi-step workflows, including conditional logic based on previous tool results.
8. Implement proper error handling with MCP error codes (InvalidRequest, MethodNotFound, InternalError) and human-readable messages helping debug integration issues.
9. Test server using MCP Inspector tool, verifying each tool responds correctly to valid inputs, rejects invalid inputs with clear errors, and handles edge cases gracefully.
10. Write client configuration examples for Claude Desktop, Claude Code, and other MCP-compatible clients with exact JSON configuration blocks ready to copy.

## Technical Standards

- Tool descriptions explain what the tool does, when to use it, what it returns—not just operation name.
- Input schemas validate all parameters before processing, returning structured validation errors.
- Resources implement both `resources/list` and `resources/read` handlers.
- Long-running operations report progress through MCP progress notifications.
- Server handles concurrent tool invocations without race conditions or shared state corruption.
- Tool results are deterministic for identical inputs unless tool explicitly interacts with external state.
- Server gracefully handles client disconnection and cleans up resources.
- Logging uses structured JSON format with request IDs for tracing tool invocations across client and server.

## Verification

- Test every tool with MCP Inspector, confirm correct responses for valid inputs.
- Send malformed requests, verify server returns proper error codes without crashing.
- Verify server starts and completes capability negotiation handshake successfully.
- Test resource listing and reading for all registered resource URI patterns.
- Confirm client configuration JSON works with at least one MCP-compatible host application.