MCP in plain words, and how to connect your first server in Claude Code
What the Model Context Protocol is, the three commands you need in Claude Code, where the config is saved, and four checks before you connect anything.
What MCP is
The Model Context Protocol (MCP) is an open standard for connecting AI applications to other systems: files, databases, calendars, your own app. The official introduction puts it this way: "Think of MCP like a USB-C port for AI applications."
Three parts, in the words of the architecture page:
| Part | What it is |
|---|---|
| Host | The AI application you work in (for example Claude Code) |
| Client | A connection the host opens to one server; one client per server |
| Server | A program that provides context: it can run locally or remotely |
A server can offer three things:
- Tools: actions the AI can run (send a query, create a draft, call an API).
- Resources: data the AI can read (file contents, records).
- Prompts: reusable templates for a task.
Connect a remote server in Claude Code
Remote servers use the HTTP transport. The basic command from the Claude Code docs:
claude mcp add --transport http <name> <url>Then, inside Claude Code, type /mcp to see the server's status and to sign in if it uses OAuth.
If the server uses a key instead of OAuth, the docs show a header option:
claude mcp add --transport http <name> <url> --header "Authorization: Bearer <token>"Keep the token in an environment variable or your OS keychain, not in a file inside a project you share.
Where the config is saved (scopes)
| Scope | Loads in | Shared with team | Stored in |
|---|---|---|---|
| Local (default) | This project only | No | ~/.claude.json |
| Project | This project only | Yes, via git | .mcp.json in the project root |
| User | All your projects | No | ~/.claude.json |
Use local while you test, project when the whole team needs the same server, user for a server you want everywhere.
Four checks before you connect a server
- Who runs it, and where? A local server runs on your machine; a remote one runs on someone else's.
- Which tools can write or delete? List them. Read-only tools are low risk; "delete", "send", "publish" are not.
- How does it sign in? OAuth lets you approve and revoke access in the browser. A static key works until someone copies it.
- Where does the key live? Environment variable or keychain, never committed to git.
Debugging
The MCP project publishes an inspector for testing servers: github.com/modelcontextprotocol/inspector. Reference server implementations live in github.com/modelcontextprotocol/servers.
If you use SocialOS
SocialOS has an MCP server, so Claude Code (or another MCP client) can read your brand kit and create post drafts, render animated carousels and, if you allow it when you connect, publish. You choose per connection: publish straight away, or queue every post for your approval in SocialOS. Sign-ups are currently by waitlist: agenticos.agency.
Sources
- modelcontextprotocol.io, "What is the Model Context Protocol (MCP)?" and "Architecture overview", checked 2026-10-10.
- Claude Code docs, "Connect Claude Code to tools via MCP" (code.claude.com/docs/en/mcp), checked 2026-10-10.
Made by SocialOS. We build a tool that plans and drafts a month of posts for small businesses; nothing is published without your approval unless you switch on autopilot.
