Plug in your first server | Master AI Automation in 4 hours Master AI Automation in 4 hours Course About Ayush Modules Sample chapter Toolbox The Microcap Minute Classroom / Module 05: MCP: Plug-ins With Standards / Chapter 3 Plug in your first server Watch first, then read. Same lesson, your pace. What you will learn – Where MCP clients store their configuration – Connecting the official filesystem server, step by step – The universal install ritual for every future server Meet the client MCP servers need a client to run inside. For this course: Claude Desktop (claude.ai/download), free, stable, the reference implementation. (OpenCode, from Module 10, speaks the same protocol from your terminal.) Install it, log in, then open its config file, a little JSON that lists connected servers: Mac: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%Claudeclaude_desktop_config.json Can’t find it? Claude Desktop → Settings → Developer → Edit Config opens it directly. The wiring Start conservative: grant access to one workspace folder, not your whole home. Add: { “mcpServers”: { “filesystem”: { “command”: “npx”, “args”: [ “-y”, “@modelcontextprotocol/server-filesystem”, “/Users/ayush/ai-workspace” ] } } } (Windows users: swap in your path, e.g. C:\Users\ayush\ai-workspace .) Read the JSON with Chapter 4 eyes: an object of servers; each has a command ( npx fetches and runs the package ) and args (the package name + which folders it may touch). Save. Fully quit Claude Desktop (not just close) and reopen. A tools icon appears near the chat box, click it: read_file , list_directory , create_file … your AI grew hands. First conversation through the door Ask: “List what’s in my ai-workspace folder.” Watch carefully, before answering, Claude shows a permission prompt: “Claude wants to use list_directory.” Approve. The listing returns, sourced from your actual disk. That prompt is the seatbelt of this whole module: nothing executes silently. You can allow once, always-for-this-chat, or deny. Keep approving consciously this week; reflexes come later, vigilance comes first. Now push further: “Create a file called hello-mcp.txt containing today’s date” → watch create_file fire “Read it back to me” → two tools chained in one reply “Rename it to day1.txt” → done, no Finder involved You’re doing real work through natural language while every action shows its face. This is exactly how OpenCode will operate on entire codebases in Module 10, same socket, bigger muscles. The ritual you now know Every future server follows these identical steps, this chapter taught a template, not one install: Find the package/docs (Chapter 2’s vetting first) Add block under mcpServers in the config JSON Save → full restart → check the tools icon Test one benign task; watch the permission prompts Keep if useful, delete the block if not Try it yourself Complete the wiring above against your real ai-workspace . Then run the three tasks listed (create/read/rename). Finally break something small on purpose: misspell the folder path in config, restart, observe the server fail to appear, fix it. Debugging config is half of real-world MCP use, better to learn it on purpose. Key takeaways – Clients read a JSON config listing servers; Claude Desktop’s is the reference spot. – Scope early servers narrowly, one folder, not your whole machine. – Permission prompts are the seatbelt: approve consciously. – Install ritual: vet → wire → restart → verify → test → keep or cut. Download the exercise sheet (PDF) Module workbook (PDF) ← Prev: Under the hood Next: The shelf → Classroom / Module 05: MCP: Plug-ins With Standards / Chapter 3 Plug in your first server What you will learn – Where MCP clients store their configuration – Connecting the official filesystem server, step by step – The universal install ritual for every future server Meet the client MCP servers need a client to run inside. For this course: Claude Desktop (claude.ai/download), free, sta
Plug in your first server
Written by
in