build-mcp-server

Guides developers through designing and scaffolding Model Context Protocol servers for Claude.

16|1|Updated Mar 20, 2026
One-click install
npx skills add https://github.com/SocialGouv/iterion --skill build-mcp-server-socialgouv
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: build-mcp-server
Source: https://github.com/SocialGouv/iterion/tree/main/vendor/github.com/SocialGouv/claw-code-go/internal/tools/bundled_skills/mcp-server-dev/skills/build-mcp-server
Command: npx skills add https://github.com/SocialGouv/iterion --skill build-mcp-server-socialgouv

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Building an MCP server involves early decisions — deployment model, tool-design pattern, framework, and auth — that are painful to reverse if chosen wrong. This Skill structures that discovery process before any code is written, then scaffolds or hands off to specialized skills. ## Core Features & Use Cases - Use-case interrogation: Asks targeted questions about what the server connects to, who uses it, action count, UI needs, and upstream auth to determine the right architecture. - Deployment model recommendation: Chooses between remote streamable-HTTP servers, MCP apps with UI widgets, MCPB bundled local servers, or local stdio prototypes. - Tool-design patterns: Applies one-tool-per-action for small API surfaces or the search + execute pattern for large APIs with dozens of endpoints. - Use Case: A developer wants to wrap a SaaS API with 60 endpoints for Claude users. The Skill recommends a remote HTTP server with the search + execute pattern, scaffolds it with the TypeScript SDK or FastMCP, and walks through OAuth via CIMD and Anthropic Directory submission. ## Quick Start Ask the assistant to help you build an MCP server that wraps your API and answer its discovery questions about deployment and auth.

Frequently Asked Questions about build-mcp-server

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I build an MCP server for Claude?▼

Start by deciding the deployment model: remote streamable-HTTP is recommended for cloud APIs, MCPB for local machine access. Then pick a tool pattern, choose the TypeScript SDK or FastMCP for Python, scaffold the server, and test it as a custom connector in Claude.

Should I use one tool per action or search and execute for my MCP server?▼

Use one tool per action when exposing under roughly 15 operations so Claude sees the full surface directly. For large APIs with dozens of endpoints, expose search_actions and execute_action tools to keep tool schemas from flooding the context window.

TypeScript SDK vs FastMCP for MCP server development?▼

The official TypeScript SDK (@modelcontextprotocol/sdk) has the best spec coverage and gets new features first. FastMCP 3.x is the choice when you prefer Python or are wrapping a Python library, offering decorator-based low-boilerplate tool definitions.

Does MCP support OAuth authentication for remote servers?▼

Yes, remote MCP servers support OAuth 2.0 via CIMD (Client ID Metadata Document, preferred in spec 2025-11-25) or DCR as a fallback. Claude supports oauth_dcr, oauth_cimd, and Anthropic-held credentials, but not user-pasted static bearer tokens.

When should I use elicitation instead of MCP app widgets?▼

Use elicitation for simple structured input like confirmations, enum pickers, or short flat forms — it is spec-native with zero UI code. Escalate to widgets when you need nested data, searchable lists, visual previews, or live updates, and always check client capability first.

Why is local stdio not recommended for distributing MCP servers?▼

Local stdio servers require users to have the right Node or Python runtime installed, cannot receive pushed updates, and only distribute through Claude Code plugins. For local servers that must ship to users, MCPB packaging bundles the runtime instead.