What Is MCP? How Creators Can Connect Tools and Libraries to Claude
MCP, the Model Context Protocol, is an open standard that lets AI apps such as Claude connect to outside tools and data through programs called MCP servers. You add a remote server as a custom connector in the Claude apps, or with claude mcp add in Claude Code, and then sign in. From then on, Claude can use the tools that server offers.
What MCP is, in plain English
MCP is a shared connection standard that lets Claude reach tools and data outside the chat. The name stands for Model Context Protocol. Anthropic introduced it in November 2024 and released it as open source, and its specification, software kits and guides are published at modelcontextprotocol.io. Many AI apps and developer tools support it, not only Claude.
The project’s docs compare MCP to a USB-C port for AI apps. Before a shared plug existed, every device needed its own cable. Before MCP, every AI app needed its own custom link to every service. With one shared format, a service can build one MCP server and any app that speaks MCP can plug into it.
Three roles are worth knowing:
- The host is the AI app you use, such as Claude Desktop or Claude Code.
- The server is a program that offers something to that app. It can run on your own computer (a local server) or on the internet (a remote server).
- The client is the connection the host opens to each server, one per server. You rarely see it.
A server can offer three kinds of things. Tools are actions Claude can call, such as searching a catalogue or creating a file. Resources are data Claude can read, such as a document or a database record. Prompts are reusable templates the server provides. Claude picks a tool when your request calls for it, and the app passes the call to the right server and hands the answer back.
How to add a remote MCP server to Claude
Adding a remote server takes its web address and a sign-in, and the steps differ slightly between the Claude apps and Claude Code.
In the Claude apps (claude.ai, Claude Desktop, the mobile apps and Cowork), a remote server is added as a custom connector. According to the Claude Help Center at the time of writing, an individual user:
- Opens Customize > Connectors, clicks + Add, then Add custom connector.
- Names the connector, pastes the server’s URL and clicks Continue.
- Reviews the sign-in settings the app detects, picks when to sign in, and clicks Add.
To use it, open the + menu in a chat, choose Connectors and switch it on for that conversation. On Team and Enterprise plans, an owner adds the connector for the organisation first, and each member then connects their own account.
The Help Center currently lists custom connectors on Free, Pro, Max, Team and Enterprise plans, with Free accounts limited to one. Plan rules change, so check that article before you rely on them. Because the connection comes from Anthropic’s cloud, the server has to be reachable from the public internet.
In Claude Code, you add a remote server from your terminal. The docs recommend the HTTP transport for remote servers:
claude mcp add --transport http <name> <url>Replace <name> with a short label and <url> with the address the provider gives you. If the server needs a sign-in, start Claude Code, type /mcp and follow the login steps in your browser.
By default the server is added for the current project only. To use it in every project on your computer, add the user scope:
claude mcp add --transport http <name> --scope user <url>claude mcp list shows what is connected, and claude mcp remove <name> takes a server away. If you log in to Claude Code with your claude.ai account, connectors you added in the Claude apps appear in Claude Code automatically.
What a library over MCP does that copy and paste cannot
A library served over MCP becomes something Claude can search and pull from directly, instead of a page you copy from. Copy and paste works well, and it is how the free effect prompts on this site are used today: you find a prompt, copy it, fill in the blanks and paste it into Claude. Claude sees only what you paste, and if the prompt is improved later, your copy stays as it was.
Over MCP, a library’s server can offer tools of its own. One tool might search the library by keyword. Another might fetch an item’s full text into the conversation. In Claude Code, which can write files on your computer, Claude can then install a fetched skill by saving it into your skills folder, ready for the next edit. You ask in plain English, such as “find a caption style for a fast tutorial and install it”, and Claude chains the steps, stopping for your approval where your settings require it.
Claude Code also supports the other two building blocks. You can point Claude at a server’s resources with @, the same way you mention a file, and a server’s prompts show up as commands in the / menu.
| Task | Copy and paste from a website | Library connected over MCP |
|---|---|---|
| Finding an item | You browse and search the site yourself | Claude searches the library from inside the chat |
| Getting the full text | You copy it and paste it into Claude | Claude fetches it into the conversation |
| Installing a skill | You download files and move them into the right folder | In Claude Code, Claude saves the files to your skills folder |
| Staying current | Your pasted copy never changes | Each fetch returns what the server holds now |
| Setup | None | Add the server once and sign in |
AI Creator Library plans to offer MCP access with its paid plans, which are listed on the pricing page. It is not live yet, so for now everything on the site works by copy and paste.
Security basics before you connect a server
Connect only servers you trust, because a server’s tools act with whatever access you grant them. Anthropic does not verify custom connectors, and the Claude Code docs carry the same warning: check that you trust each server before connecting it.
- Know who runs it. Use servers built and hosted by companies or developers you trust. A server’s owner can change what its tools do without telling you.
- Sign in with OAuth. Most remote servers use OAuth, the familiar browser screen where you sign in to the service and approve access. Claude never sees your password, and you can revoke access later from Claude’s settings or from the service itself.
- Read what you approve. The sign-in screen lists the permissions the server wants. Grant the narrowest access that does the job.
- Watch tool approvals. When the Claude apps ask to run a connector’s tool, read the request before you approve it. Choose “Allow always” only for tools you trust to run unsupervised, and switch off tools you do not need, especially ones that write or delete. In Research, Claude can call connector tools without asking, which is one more reason to turn off write tools you do not need. In Claude Code, permission rules can allow or block a whole server’s tools.
- Expect hidden instructions. A malicious server, or an honest one that reads content from the web, can return text that tries to steer Claude. This is called prompt injection. Keep an eye on what tools send and receive when your files or accounts are involved.
Claude Code adds one more guard: it asks for approval before using servers listed in a project’s shared .mcp.json file, so a project you download cannot quietly connect itself to a server.
How MCP fits with skills and prompts
MCP connects Claude to things, while skills and prompts tell Claude what to do with them. The Claude Help Center draws the same line: MCP links Claude to outside services and data, and a skill carries the procedure for a task. A prompt is the lightest of the three, a set of instructions for one conversation.
For a creator, the pieces stack. A library connector finds and fetches material, an editing skill holds your style, and a short prompt says what this particular video needs. If you have not set up Claude Code for video yet, the one-time setup is the place to start, and what a Claude skill is explains the skill side in more detail.
Where this falls short
- For a prompt you will use once, copy and paste is quicker than adding a server and signing in.
- Custom connectors in the Claude apps are reached from Anthropic's cloud, so a server that runs only on your computer or a private network will not connect there without extra setup. A local server in Claude Code suits that case better.
- A remote server can change its tools or go offline without notice, while a prompt or skill saved on your computer stays exactly as you left it.
- On Team and Enterprise plans, members cannot add a custom connector alone; an owner or admin has to add it for the organisation first.
Questions
Is MCP only for Claude?
No. Anthropic created MCP and released it as an open standard, so any AI app can support it, and the project's documentation lists a range of AI assistants and code editors that already do. One MCP server can therefore work with several apps, although each app has its own way of adding a server and its own plan rules.
What is the difference between a connector and an MCP server?
In the Claude apps, connector is the name for an MCP server added to your account. A custom connector is one you add yourself by pasting a remote server's URL, as opposed to choosing a reviewed one from Anthropic's directory. Both run on MCP, and Claude Code's docs note that any remote server listed in that directory can also be added with claude mcp add.
Do I need to write code to use an MCP server?
No. Using a remote server means pasting a URL into the Claude app's connector settings, or running one command in Claude Code, then signing in. Code only comes in if you want to build a server of your own. Local servers that run on your computer can need a little more setup, such as a runtime the server depends on.
Can I disconnect a server after adding it?
Yes. In the Claude apps you can switch a connector off for a single chat from the + menu, and you can revoke its access at any time from Claude's settings or from the connected service's security settings. In Claude Code, claude mcp remove followed by the server's name deletes it, and the /mcp menu can clear its stored sign-in.
Read next
Sources (6)
- What is the Model Context Protocol (MCP)? — Model Context Protocol. Accessed October 21, 2026.
- Architecture overview — Model Context Protocol. Accessed October 21, 2026.
- Introducing the Model Context Protocol — Anthropic. Accessed October 21, 2026.
- Get started with custom connectors using remote MCP — Claude Help Center. Accessed October 21, 2026.
- Connect Claude Code to tools via MCP — Anthropic — Claude Code docs. Accessed October 21, 2026.
- What are skills? — Claude Help Center. Accessed October 21, 2026.