Skip to main content
Any client that speaks MCP can use Context. There are two ways in.

​
Local, through the CLI

Install and sign in once, then let the CLI add Nemu to your editor:
Setup writes only the nemu entry. Every other server, setting and comment in the file, and its indentation, stay exactly as they were, and the previous file is kept next to it with a .nemu-backup suffix. A file that cannot be parsed is left alone, and so is a different server that is already called nemu. Restart the editor afterwards. nemu uninstall <editor> takes the entry out again, and nemu status lists the editors it finds. Hooks, which load notes when a session starts, exist for Claude Code and Codex only. The entry starts nemu mcp, which runs on your machine, uses your sign in, refreshes it, and knows which project you are in from the folder the client started in. For any other client, nemu mcp config <client> prints the snippet and where it goes, for cursor, gemini, vscode, windsurf, zed, opencode, claude, codex and remote. For Cursor it is:
~/.cursor/mcp.json
OpenCode keeps the program and its arguments in one command array:
~/.config/opencode/opencode.json
Run opencode mcp list afterwards to see it connected.

​
Remote, over HTTP

The server is at https://app.nemu.cc/api/mcp (Streamable HTTP). Authenticate in either of two ways. Sign in with OAuth. Clients that support MCP authorization, such as Claude Code’s /mcp, claude.ai custom connectors and ChatGPT connectors, discover Nemu’s sign in on their own: add the URL and approve the request in your browser. The app registers itself and can only read and write Context. It cannot reach the rest of your account. Use a Context key. Create one in the console under Context, Keys and send it as a bearer token.
A remote client does not know which folder you are in, so the agent opens the project itself with the project_open tool, passing the repository name and its first commit. See MCP.