MCP Server Setup: A Practical Guide

What an MCP server actually is

MCP — the Model Context Protocol — is the standard Claude Code uses to connect to things outside the conversation: a database, a filesystem, an external API. An MCP server is a small program that exposes one of those over that protocol, and Claude Code reads a JSON configuration at startup to know which servers exist and how to reach them.

The mcp.json schema

The configuration is a single JSON object with one required top-level key: mcpServers, whose value is an object keyed by server name. Each entry names the command to run the server and its arguments; any credential it needs goes in an env block, referenced as a placeholder, never written in directly.

{
  "mcpServers": {}
}

That is a completely valid configuration — an empty mcpServers object. There is no requirement to configure anything; a server only gets an entry once you actually need it.

Setting one up, step by step

1. Confirm the server you need — the tool or data source you want Claude Code to reach, and the MCP server that exposes it.
2. Add an entry under mcpServers with the command and args needed to run that server.
3. Reference any secret the server needs as an ${ENV_VAR} placeholder in an env block — never hardcode it.
4. Reload Claude Code so it picks up the updated configuration.
5. Verify the server shows as connected and a simple request against it succeeds before relying on it.

A worked example

For a project that reaches the local filesystem, a Postgres database, and GitHub, the configuration might look like this — three servers is an example, not a default; your own file should only list what your project actually needs.

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@mcp/server-filesystem", "/workspace"]
    },
    "postgres": {
      "command": "node",
      "args": ["./mcp/pg-server.js"],
      "env": { "DATABASE_URL": "${DATABASE_URL}" }
    },
    "github": {
      "command": "npx",
      "args": ["-y", "@mcp/server-github"]
    }
  }
}

Project-level vs global config

A project-level .claude/mcp.json is scoped to that repository only — it travels with the project and applies whenever Claude Code runs there. A separate global configuration can make a server available across every project, but merging a project-level setup into that global file should always be an explicit, opt-in action you take deliberately — never something that happens silently as a side effect of setting up one project.

How Logic64 fits

Logic64 generates .claude/mcp.json as part of the governed workspace it produces from your locked architecture. It populates the file from what you actually configure during planning — and never invents a server you did not ask for. If nothing is configured, it emits a valid, empty mcpServers object rather than guessing.

When you run logic64 pull, that file is routed and safely merged into your local project alongside the rest of the workspace — not silently pushed into your global configuration.

Frequently asked questions