innernet · mcp
mcp servers vs ai skills, and where to find both
one gives your ai access, the other teaches it judgement. most people need both, and they are found in different places.
"Where do I find MCP skills?" is one of the most common questions people ask once they've connected their first tool, and it is two questions folded together. MCP servers and skills both make an AI more capable. They do it in opposite ways, and they are found in different places.
the short version
An MCP server gives an AI access. It is running software that connects the AI to a system (your files, a database, a calendar, a design tool) and exposes things it can read or do there.
A skill gives an AI judgement. It is a document that teaches the AI how to do one job well: the steps, the rules, what to check.
A server without a skill is an AI that can reach your database but doesn't know your conventions. A skill without a server is an AI that knows exactly how to review a pull request but can't see the pull request. The useful setups have both.
what an MCP server is
The Model Context Protocol is an open standard for connecting AI applications to outside systems. A server speaks it on one side and talks to a system on the other. Once an AI client is connected, the server offers:
- tools, actions the AI can call (create an issue, run a query, send a message);
- resources, things it can read (a file, a record, a page);
- prompts, ready-made instructions the server suggests.
Servers run either on your machine or remotely, behind a sign-in. Connecting one is usually a line of configuration or a click in your AI tool's connector settings.
what a skill is
A skill is a folder with a SKILL.md file: a name, a description of when to use it, and instructions. The AI reads the description of every skill it has, and opens a skill in full only when a request matches. Nothing runs. It is a document, which means it costs nothing to keep around and works in any tool that reads it. What is a SKILL.md covers the format.
when you need which
| you want your ai to... | you need |
|---|---|
| read or change something in another system | an MCP server |
| do a job the way you or your team does it | a skill |
| follow a house style, method or checklist | a skill |
| act on live data without you pasting it in | an MCP server |
| use a system correctly, not just reach it | both |
The last row is the common one. Database companies now publish a server that connects to their product and a skill that teaches the right way to use it. Supabase's postgres best practices is a skill of that kind; Anthropic's mcp builder is a skill for writing servers.
where to find mcp servers
The official MCP registry. The protocol's maintainers run a public registry that servers publish to under a verified namespace. It is the closest thing to a canonical list, and the namespace tells you who published a server.
The company whose product you use. Most products that matter now ship an official server. Prefer it over a third-party one for anything that touches your account.
Your AI tool's connector directory. Claude, ChatGPT and several editors keep a directory of servers they have reviewed. Fewer choices, less checking to do.
Community lists on GitHub. Long, useful for discovering what exists, and uneven in quality.
Before connecting a server, check who publishes it, what it can do rather than only read, and whether it signs in with your account or asks you to paste a key. A server acts with whatever access you give it.
where to find skills
Official repositories from the companies that build AI tools, which double as reference examples of the format.
Company repos from infrastructure makers, for using their products correctly.
Community collections from individual makers, where the most distinctive methods live.
innernet's community, which keeps all three kinds as links rather than downloads, each with its author, its source and when it last changed. Where to find ai skills walks through each.
Before installing a skill, read it. It is short, and it is instructions your AI will follow with your access.
innernet is both
innernet is an MCP server: connect it once and every AI tool you use reads and writes the same memory of your projects and decisions. The skills people keep there are served through that same connection, so a skill you keep follows you into every tool that is connected, at the version its author keeps now.
questions
what is the difference between an mcp server and a skill?
An MCP server is running software that connects an AI to an outside system so it can read data or take actions there. A skill is a markdown document that teaches an AI how to do a job. A server gives access; a skill gives method.
where can i find mcp servers?
The official MCP registry, the company whose product you want to connect, your AI tool's own connector directory, and community lists on GitHub. Prefer an official server from the product's maker for anything that touches your account.
where can i find mcp skills?
"MCP skills" usually means skills, which are separate from MCP. Find them in official skill repositories from AI companies, skills published by infrastructure companies, community collections, and innernet's community page, which keeps them as links with the author and last change on each.
do i need both mcp servers and skills?
Often, yes. A server lets your AI reach a system and a skill teaches it to use that system correctly. Many companies now publish both for the same product.
innernet is user-owned memory across AI tools, delivered as one MCP server. Your projects, your decisions and your skills, read by every tool you connect. innernet.live
Checked 17 september 2026.
keep reading
one memory, every ai tool.
innernet keeps your projects, your decisions and your skills in one place, and every ai tool you connect reads the same thing. the skills people keep there are open to anyone.