Connect Codex, Cursor and Gemini CLI · Academy

Who it is for: people who use another AI coding tool instead of (or next to) Claude Code. What you need: a Nexorah account with an active plan, Python 3, and the tool itself.

Honest status: the plugin script below was tested against the reference server. The MCP settings of Option B follow each tool's official documentation (checked 2026-11-08, sources at the end) but have not been tried against the live service.

Option A: the plugin script (works in any tool that can run a terminal command)

Get the script from the Connect page of your account (direct file address: to be confirmed; it is not published yet). It is one Python file, nl.py, with no dependencies.
Connect this device:
```
python3 nl.py login
```
Open the printed link, sign in to Nexorah, enter the code. You see Connected. This device now uses Nexorah Library.
Tell your tool to use it. Add one instruction to the tool's project instruction file (for example AGENTS.md): "Before choosing any tool, library or method, run python3 nl.py search "<words>" and python3 nl.py card <slug>."
Ask your tool to run python3 nl.py me.

Done means: the tool runs nl.py me and shows your plan.

Option B: MCP connection

Use the MCP URL from the Connect page in place of <MCP URL>. Each tool signs you in to Nexorah through its own OAuth flow.

Cursor: put the server in ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (this project):

``json { "mcpServers": { "nexorah": { "url": "<MCP URL>" } } } `` Cursor supports OAuth for remote servers, so it asks you to sign in with Nexorah. Servers are listed, and can be switched on or off, under Customize in Cursor's sidebar.

Codex (CLI, IDE extension and ChatGPT desktop app share one configuration): add this to ~/.codex/config.toml:

``toml [mcp_servers.nexorah] url = "<MCP URL>" ` then run codex mcp login nexorah and sign in with Nexorah. On a machine without a browser add --no-browser, open the printed link elsewhere and paste the final redirect link back into the terminal. In the ChatGPT desktop app (Settings → MCP servers) or the IDE extension (gear menu → MCP servers) you can instead click Add server, choose Streamable HTTP, enter the URL, save, restart when asked, then click Authenticate. In the codex terminal, /mcp` lists connected servers.

Gemini CLI: run

``bash gemini mcp add --transport http nexorah <MCP URL> ` (add --scope user to have it in every project; the default is this project only), start gemini, then run /mcp auth nexorah` to sign in. Account note: Google ended Gemini CLI for Google AI Pro, Ultra and free individual use on 18 June 2026 (Antigravity CLI replaced it). Organisation access with a Gemini Code Assist Standard or Enterprise licence, and paid API keys, stay supported. If your account is not covered, use Option A in whichever tool you use.

First conversation: the start wizard

Once connected, you do not have to learn the library's tools. Ask your AI to set you up with Nexorah Library, or run the start command if your app lists it: with the Nexorah plugin in Claude Code it is /nexorah-library:start; with a plain MCP connection it appears as /mcp__nexorah__start. Your AI then calls the nexorah_start tool and walks you through four short steps:

your setup in one line: plan, sharing choice, today's question count;
three questions about your stack, what you build, and what your AI should always or never do; the answers are saved as My preferences only after you say yes;
a first real answer from the library for your kind of work;
the next step for the app you use.

Until you run it once, library answers begin with the line "New here: call nexorah_start". Claude's help pages do not say whether the Claude apps list a connector's commands, so if you do not see start there, ask "Set me up with Nexorah Library"; it does the same. Your account page shows the connection within about thirty seconds: Your connections → Connected · just now.

Troubleshooting

The tool cannot reach the server: check your network, then see guide 9 (offline).
device_limit: one device per subscription (guide 9).

FAQ

Which option is better? Option A is verified against the reference server and works everywhere. Option B is more convenient, but has not been tried against the live service yet.
Do I need a separate subscription per tool? No. One subscription, one device, any number of tools on it.

Sources (✅ checked 2026-11-08)

Cursor docs, "Model Context Protocol (MCP)": https://cursor.com/docs/mcp
Codex docs, "Model Context Protocol" (config.toml, codex mcp login, desktop and IDE steps): https://learn.chatgpt.com/docs/extend/mcp (the old address https://developers.openai.com/codex/mcp redirects here)
Gemini CLI docs, "MCP servers with Gemini CLI" (gemini mcp add, /mcp auth): https://geminicli.com/docs/tools/mcp-server/
Google Developers Blog, "Transitioning Gemini CLI to Antigravity CLI" (19 May 2026; consumer cut-off 18 June 2026; enterprise licences and paid API keys unchanged): https://developers.googleblog.com/an-important-update-transitioning-gemini-cli-to-antigravity-cli