| name | mcp-server-typescript |
| description | Build MCP servers in TypeScript using the official SDK. Exposing internal functions as tools, prompts, resources. Covers transport (stdio, Streamable HTTP), auth (API keys, OAuth 2.1), debugging with MCP Inspector, and remote exposure via Tailscale/Cloudflare Tunnel. Use when building, scaffolding, or deploying MCP servers in TypeScript/Node. |
MCP Server — TypeScript
Build MCP servers using the official @modelcontextprotocol/sdk (v1) or @modelcontextprotocol/server (v2).
SDK versions
Two generations coexist. Choose based on stability needs:
| v1 (stable) | v2 (current) |
|---|
| Package | @modelcontextprotocol/sdk | @modelcontextprotocol/server |
| Import | @modelcontextprotocol/sdk/server/mcp.js | @modelcontextprotocol/server |
| Zod | zod (v3.25+) | zod/v4 (or any Standard Schema) |
| Schema | Zod only | Standard Schema (Zod v4, Valibot, ArkType) |
| Status | Bug fixes, security patches | Active development |
v1 imports use deep paths (/sdk/server/mcp.js). v2 uses top-level package names. All examples in this skill use v1 unless noted — it's what most production servers run today. Migration is straightforward: change imports, switch to zod/v4.
Primitives
- Tools — LLM-invoked actions with validated input/output schemas. See TOOLS.md.
- Resources — URI-addressed read-only data (files, DB records, API responses). See RESOURCES.md.
- Prompts — Reusable message templates with typed arguments. See PROMPTS.md.
Transport
- stdio — local servers spawned as child processes (Claude Desktop, CLI tools)
- Streamable HTTP — remote servers over network (recommended for multi-client)
- SSE — legacy, avoid for new projects
Details and session management: TRANSPORT.md.
Auth
Progresses from simple to production-grade:
- Environment variable — API key in env, validated at startup
- Bearer token — static token checked per-request
- OAuth 2.1 + PKCE — full spec compliance for public remote servers
Details: AUTH.md.
Debugging
- MCP Inspector — browser UI for testing tools, resources, prompts
- Claude Desktop —
developer_mode log tailing
- Server-side logging — stderr for stdio, structured logs for HTTP
Details: DEBUGGING.md.
Remote exposure
Make a local MCP server reachable from remote clients (Claude.ai, ChatGPT, other agents):
- Tailscale Serve/Funnel — private mesh or public HTTPS, zero config
- Cloudflare Tunnel — public endpoint with Zero Trust access policies
- Direct deploy — Cloudflare Workers, Fly.io, Railway
Details: REMOTE.md.
Connecting to Claude
Claude.ai supports remote MCP servers via the integrations UI. Also covers Claude Desktop claude_desktop_config.json for local stdio servers. See CONNECTING.md.
Project structure & quality checklist
Scaffolding, package.json, tsconfig.json, build pipeline, and pre-ship checklist: PROJECT.md.
