The field guide for Claude & Cursor
Build Skills, Agents
and MCP servers.
Step-by-step guides, templates and checklists for each stage, from the first idea to a tested, shipped build.
- Track 01SkillsPackage know-how the agent loads only when it's needed.
- Track 02AgentsSubagents and SDK agents that take on focused jobs.
- Track 03MCP serversConnect agents to your tools and data, safely.
A working reference for designing, building, testing, and shipping the three main ways to extend AI coding assistants such as Claude (Claude Code, Claude.ai, Claude Desktop, Claude API / Agent SDK) and Cursor (IDE, CLI, SDK, Cloud Agents).
Each section has four kinds of document:
- Concepts, which explain what the thing is and when to use it.
- Step-by-step guides, which are numbered phases. Every step has a Goal, what to Do, and how to Verify it.
- Checklists, which you copy into a PR or ticket and tick off as you go.
- Templates, which are copy-and-paste starting points that pass the automated QA.
Start here
Section titled “Start here”| If you want to… | Read |
|---|---|
| Decide whether you need a skill, an agent, an MCP server, or just a rule | Choosing the right tool |
| Know where files go in Claude and in Cursor | Platform matrix |
| Look up a term | Glossary |
| Follow the full lifecycle for any of the three | Universal build lifecycle |
1. Skills: packaged know-how the agent loads on demand
Section titled “1. Skills: packaged know-how the agent loads on demand”| Doc | Purpose |
|---|---|
| Concepts | Anatomy, progressive disclosure, invocation |
| Step-by-step: build a skill | From idea to installed skill |
| Writing guide | Descriptions, instructions, degrees of freedom, patterns |
| Testing & evaluation | Trigger tests, eval sets, iteration |
| Distribution | Personal, project, team, plugin, Claude.ai, API |
| Checklist | Gate for every phase |
| Templates | Basic skill, skill with scripts and references |
2. Agents: subagents and programmatic agents
Section titled “2. Agents: subagents and programmatic agents”| Doc | Purpose |
|---|---|
| Concepts | Subagents, SDK agents, AGENTS.md, orchestration |
| Step-by-step: build a subagent | Claude Code and Cursor file-based subagents |
| Step-by-step: build an SDK agent | Claude Agent SDK and Cursor SDK |
| Design patterns | Orchestrator/worker, verifier, pipelines, guardrails |
| Testing & evaluation | Delegation tests, golden tasks, regression |
| Checklist | Gate for every phase |
| Templates | Subagent files, SDK scripts |
3. MCP servers: connect the agent to tools and data
Section titled “3. MCP servers: connect the agent to tools and data”| Doc | Purpose |
|---|---|
| Concepts | Tools, resources, prompts, transports, lifecycle |
| Step-by-step: TypeScript server | Build a stdio server with the official TS SDK |
| Step-by-step: Python server | Build a server with the Python SDK (MCPServer) |
| Connect to clients | Cursor, Claude Code, Claude Desktop, Claude.ai |
| Tool design | Naming, schemas, outputs, errors, token budget |
| Testing & debugging | Inspector, logs, automated tests |
| Security & production | Auth, secrets, remote deploy, versioning |
| Checklist | Gate for every phase |
| Templates | TS server, Python server, client configs |
Master checklist
Section titled “Master checklist”MASTER-CHECKLIST.md is a one-page “definition of done” covering all three.