Use casesWorks with

Give Claude Code a memory that lasts between sessions

Claude Code connects to Context Engine as a remote MCP server with one claude mcp add command and a sign-in in the browser; after that it reads a brief at the start of each session and saves decisions, notes and tasks as it works.

What it does

Claude Code forgets everything when a session ends. Connected to an engine, it starts each session with a brief of what the engine is for and what changed since it was last there, searches by meaning across every engine the connection reaches, and saves what it learns as typed memories: decisions with their reason, notes, tasks, people and files. Your teammates’ Claude Code, Cursor and ChatGPT read the same memory, and the people on the team can read it in the app.

Connect it

  1. Sign up at app.contextengine.com and make an engine.
  2. Open AI tools, then Connect an AI tool, and choose Claude Code. Under Change… pick its engines, each at Can read or Can read and save.
  3. Copy the claude mcp add command it shows and run it in your terminal. The server is named after the connection, starting with ce-.
  4. In Claude Code, run /mcp, choose that server and Authenticate. Allow it in the browser window that opens.

The screen turns Signed in green when Claude Code first reaches the connection, then First memory when it saves something. Show me how explains --scope and the project variant.

No browser where it runs?

For Claude Code on a server or in a container, choose Use a key instead under the steps. It makes a personal key named after the app and shows it once, already inside the command, pointed at the shared https://app.contextengine.com/mcp address.

Who can see what it saves

Everyone in the engine sees the memory as “Claude Code” followed by whose it is. Owners and admins see every connection reaching their engine and can cut it off; anyone can pause their own, and every change records who made it, with which tool, and what it was before.

Questions

Does Claude Code need an API key for Context Engine?

No. It signs in through the browser with OAuth. A key is only for agents that run where no browser can open, such as a server or a container; Connect an AI tool offers "Use a key instead" for that.

How do I keep two clients' memories apart in Claude Code?

Use a project-scoped connection per client repository. Each connection reaches only the engines you tick, and in Claude Code its server is named after the connection, so two can sit side by side.

What does Claude Code see when a session starts?

A brief for each engine the connection reaches: what the engine is for, what happened recently and what changed since it was last there.

In the help