Claude Code MCP server that remembers your notes

What a Claude Code MCP server is

An MCP server is a separate program that speaks the Model Context Protocol, the open standard that Claude Code, Anthropic's command-line coding agent, uses to connect outside tools. Each server offers a few named tools. Claude Code reads their names and descriptions, then decides when to call one, the same way it decides to run a shell command or open a file.

Some servers run on your computer and talk to Claude Code over stdio, which means Claude Code starts the program and swaps messages with it directly. Others are hosted, and Claude Code reaches them over HTTP, usually after you sign in once. Locul offers both: a local server that reads your notes from disk, and a hosted connector for machines where Locul isn't installed.

Anthropic's Claude Code MCP documentation is the reference for every flag below, and the Model Context Protocol site explains the protocol itself. Every command on this page was checked against those docs on 16 October 2026.

How to add an MCP server to Claude Code

Claude Code adds every MCP server with one command, claude mcp add, and the flags change with how the server runs. To add a hosted server, run claude mcp add --transport http <name> <url>. For one that runs on your computer, put its command after two dashes. Then run claude mcp list to confirm it connected. Pick the tab that matches the setup instructions you were given.

claude mcp add --transport http notion https://mcp.notion.com/mcp

claude mcp add --env AIRTABLE_API_KEY=YOUR_KEY --transport stdio airtable -- npx -y airtable-mcp-server

claude mcp add-json my-server '{"type":"http","url":"https://mcp.example.com/mcp","headers":{"Authorization":"Bearer YOUR_TOKEN"}}'

No Claude Code yet?

Anthropic's installer puts the claude command on your computer. Open a new terminal afterwards and run claude --version to check it.

macOS, Linux and WSL

curl -fsSL https://claude.ai/install.sh | bash

Windows PowerShell

irm https://claude.ai/install.ps1 | iex

Local, project or user scope

A Claude Code MCP server's scope decides which folders can use it, and a wrong scope is the most common reason a server seems to disappear: it was added with the default scope in a different folder.

When the same name exists in more than one scope, Claude Code uses the local entry first, then project, then user. Project servers from a .mcp.json ask for your approval the first time, and claude mcp reset-project-choices clears those answers.

.mcp.json in your project root

{
  "mcpServers": {
    "api-server": {
      "type": "http",
      "url": "${API_BASE_URL:-https://api.example.com}/mcp",
      "headers": { "Authorization": "Bearer ${API_KEY}" }
    }
  }
}

A .mcp.json file can read environment variables with ${VAR} or ${VAR:-default}, so you can commit it without committing your keys.

Check it works

Four commands cover almost every Claude Code MCP check. Run them in your terminal, except /mcp, which you type inside a Claude Code session.

Every server and whether it connected.

claude mcp list

One server's details, including its scope and command.

claude mcp get locul

Inside a session: status, sign-in, reconnect.

/mcp

Removes a server and any sign-in token it stored.

claude mcp remove locul

Then ask Claude Code something only the server can answer. "Which Locul tools can you see?" is a good first question.

MCP servers to add to Claude Code first

Most Claude Code setups need three or four MCP servers, not thirty. These come up again and again, each with the command from its maker's own docs. Every server costs a little context and a little trust, so add one when you catch yourself doing its job by hand.

If an add-json command fails in a Windows terminal, run it from Git Bash or WSL, where the single quotes around the JSON work as written.

The official Model Context Protocol servers repository lists the reference servers and links to hundreds more. Our ranked list of MCP servers covers each pick in depth, with what it costs in context and the popular ones you can skip.

On a Hacker News thread about must-add servers for Claude Code, one reply was "Context7 is quite useful for coding", and another user pairs a Brave Search server with the Fetch server for fast web lookups. The same thread has a fair pushback: Claude Code already searches the web and reads pages on its own, which is why the list above stays short.

Only add servers you trust

Anthropic's documentation says to verify you trust each server before you connect it, because servers that fetch outside content can expose you to prompt injection. Take servers from their makers, give each token the narrowest scope that works, and approve a project's .mcp.json servers only when you know what they run.

Troubleshooting

These are the problems people hit most often after running claude mcp add, with the fix for each.

Add Locul to Claude Code

Locul is a desktop app for Mac and Windows that turns your markdown notes, PDFs and dictations into memories your AI apps can recall, and keeps them current as your work changes. It connects to Claude Code as an MCP server called locul.

1. Install Locul and let it index your home folder. It reads markdown files, and saves a .md copy next to each PDF so those get read too. New notes are picked up while the app runs.
2. Open Connections in Locul and click Connect on the Claude Code card. Locul runs claude mcp add for you at user scope, so every project can use it. It then offers to add a short block to ~/.claude/CLAUDE.md that tells Claude Code when to check your memory. You can say no.

   This is the shape of what Locul runs. The app fills in the real paths, so you never type it:

   Copy

   claude mcp add locul --scope user \
     -e LOCUL_HOME=<your home folder>/.locul \
     -- <path to locul-node> <path to locul-mcp.js>

3. Check it. Run claude mcp list and look for locul, then ask: "What do my notes say about the billing API?"

No Locul on this machine?

On a work laptop or a remote box, add the hosted connector instead and sign in once. It reaches the memories your Locul account holds, not the files on another computer.

claude mcp add --transport http locul https://team-api.locul.ai/mcp

Then run /mcp in Claude Code and pick locul to sign in. The cloud connector page lists its tools. If you already added Locul as a connector on claude.ai, Claude Code picks it up when you sign in with the same claude.ai account.

Locul doesn't need to be open for Claude Code to use it. Claude Code starts Locul's server on its own. Locul does need to run to learn from notes you write after that.

CLAUDE.md or a memory server?

Claude Code already has memory of a kind. It reads CLAUDE.md files at the start of every session, from your project and from ~/.claude. For a repo's build commands and code style, that's the right place, and Locul's free CLAUDE.md analyzer checks one for conflicting rules.

A CLAUDE.md file stops working well when the context isn't about one repo. The decision you made in a meeting, the customer who asked for a feature, the reason you dropped a library last spring: those live in your notes, and copying them into every CLAUDE.md is how the file grows until Claude Code starts ignoring parts of it.

You don't have to choose. Keep a short CLAUDE.md for the repo, and let memory hold everything else.

Watch it first

Claude Context: Stop Pasting It Into Every New Chat Locul AI

Adding MCP Servers to Claude Code: Real Examples That Work Nathan Sebhastian

Questions people ask