guide4 min read

MCP explained (without the jargon)

MCP = Model Context Protocol. It's a standard plug shape for AI agents. One agreed- upon way to give your AI tools it can call and data it can read — so any agent that speaks MCP can use any tool that speaks MCP. Think "USB-C for AI agents": build the connector once, plug it into Claude, Cursor, or anything else.

TL;DR: an MCP server is a connector that hands your agent live tools + data (your GitHub, your database, a browser, docs…). Add one with claude mcp add …. It's different from a skill: a skill is instructions (how to do something), MCP is a live connection (things to actually do/fetch).


The core idea

Without MCP, your agent only knows what's in its training data + what you paste in. With MCP, it can reach out to real systems in real time:

  • Read your open GitHub issues, then open a PR.
  • Query your production database (read-only) to answer a question.
  • Control a real browser to test a page.
  • Pull current library docs instead of guessing from memory.

An MCP server exposes those capabilities. Your agent (the client) connects to it and can now call its tools. Servers are either local (a command that runs on your machine) or remote (a URL, often behind OAuth login).


How to add one

# Remote server (a URL) — most hosted connectors:
claude mcp add --transport http <name> <url>

# Local server (a command Claude launches):
claude mcp add <name> -- npx -y @some/mcp-server

# See what's connected / remove one:
claude mcp list
claude mcp remove <name>

Add -s user to make it available in every project, or -s project to commit it to the repo (.mcp.json) and share with your team. Inside a session, /mcp shows connected servers and lets you authenticate remote ones.


Skills vs MCP (they're not the same thing)

Skill MCP server
What it is Written instructions / a playbook A live connection to tools + data
Lives as SKILL.md markdown A running server (local command or URL)
Gives the agent Know-how — how to do a task Reach — things to actually call/fetch
Example "How to write our commit messages" "Query our GitHub / DB / browser now"
Changes at runtime? No (static text) Yes (returns live data)

They stack: a skill can tell the agent how to use an MCP tool well.


Real example servers

Server Gives your agent Add it
Context7 Up-to-date docs for thousands of libraries claude mcp add --transport http context7 https://mcp.context7.com/mcp
Playwright A real browser to click/test pages claude mcp add playwright -- npx @playwright/mcp@latest
GitHub Issues, PRs, repos, code search claude mcp add --transport http github https://api.githubcopilot.com/mcp/
Filesystem Scoped read/write to a folder claude mcp add fs -- npx -y @modelcontextprotocol/server-filesystem ~/projects

(Remote ones prompt you to log in via OAuth on first use — no tokens to paste.)


Where StackScout's MCP fits

We ship the whole StackScout skill library as one MCP connector. Add it once and every StackScout skill — the repo picks, the AI-dev playbooks, the generators — lives inside your agent, always current, no copy-pasting files:

claude mcp add --transport http stackscout https://mcp.stackscout.app/mcp

One connector, the entire library, updated on our side — your agent just gains the skills. That's the whole point of MCP: build (or subscribe to) a connector once, and every agent you use gets the powers.


Where this fits

Use a skill when you're teaching your agent how you work. Reach for MCP when your agent needs to touch a real system — your code host, your data, a browser, live docs. Most good setups use both: MCP for reach, skills for know-how.


Guide by StackScout · more free AI-dev guides, skills & repo picks → stackscout.app

◎ free guides, every week

New setups like this land with each weekly drop. Get them the moment they ship:

Join the newsletter

← all guides

MCP explained (without the jargon) · StackScout