Full Pro free for 7 days, no credit card. See pricing →
← All posts
May 28, 2026 · 6 min read

How to use Vilix with OpenAI Codex

A one-line setup: add Vilix as an MCP server in Codex, sign in once, and the Codex CLI and IDE extension both build with the project context, decisions, and rules you already captured in ChatGPT, Claude, and Cursor.

OpenAI Codex is a coding agent, not a chat app, so the value of memory here is specific: when Codex writes code, it should already know your stack, your conventions, and the decisions you made in every other AI tool this week. Vilix is the layer that carries that context in. You connect it once, and both the Codex CLI and the Codex IDE extension build with your full project memory instead of guessing.

Setup is a single terminal command. Here is the whole flow.

Add Vilix to Codex with one command

Codex stores MCP servers in its own config. Register Vilix from the terminal:

codex mcp add vilix --url https://api.vilix.ai/mcp

That points Codex at the Vilix MCP endpoint over streamable HTTP. Codex supports OAuth for streamable-HTTP servers, so you do not paste an API key , you sign in. Open the Codex IDE extension, trigger sign-in for the Vilix server, and Codex opens your browser to complete the OAuth login. Approve it, return to the editor, and the connection is live.

The same command works whether you live in the terminal or the editor. That is the part most people miss, so it is worth saying plainly.

Set it up once, CLI and IDE share the config

The Codex CLI and the Codex IDE extension read the same MCP configuration (~/.codex/config.toml, or a project-level .codex/config.toml). Add Vilix once and it is available in both surfaces automatically. There is no second setup step for the extension and no per-project re-auth, your Vilix connection follows the Codex config, not the window you happen to be in.

It is not inherited from the ChatGPT app, and that is fine

A common assumption: “I already connected Vilix in the ChatGPT app, so Codex has it too.” Not quite. The connectors you add in the ChatGPT chat app and the MCP servers Codex uses are configured separately. Codex reads its own config, which is exactly why the one-line codex mcp add step above exists.

The lines are starting to blur, OpenAI’s Secure MCP Tunnel can power ChatGPT connectors, Codex, and the Responses API from a single server, and Codex has an experimental flag for ChatGPT apps and connectors. Useful to know, still moving. The dependable path today is to configure Codex directly with codex mcp add vilix, and that is what we recommend.

What you actually get once it is connected

After connecting, Vilix’s tools show up to Codex like any other MCP server. In practice that means:

  • Codex pulls the relevant past context before it acts, the architecture decisions, the constraints, the “we don’t use X here” rules you set in another tool.
  • Your projects and tasks are visible to the agent, so “pick up the billing task” means something concrete instead of a cold start.
  • Your user_rules, short personal directives like “always show the test first”, apply here too, because they travel with your account, not the tool.

Plan a feature in ChatGPT, discuss the architecture in Claude, then have Codex implement it, and Codex already knows what was decided. That is the whole point of cross-tool memory: context that moves with you between tools instead of dying in each chat window.

Frequently asked questions

How do I connect Vilix to Codex?

Run codex mcp add vilix --url https://api.vilix.ai/mcp in your terminal, then sign in from the Codex IDE extension, it opens your browser for the OAuth login. That is the entire setup.

Do I have to set it up twice for the CLI and the IDE extension?

No. They share the same MCP config (~/.codex/config.toml or a project .codex/config.toml), so adding Vilix once enables it in both.

I already added Vilix in the ChatGPT app. Does Codex inherit it?

Not by default. Codex reads its own MCP configuration, so you still run the codex mcp add command. OpenAI is rolling out ways to share one server across ChatGPT, Codex, and the API, but configuring Codex directly is the reliable path today.

Does Codex need an API key for Vilix?

No. Vilix uses OAuth over streamable HTTP, so you sign in rather than paste a key. You can revoke access at any time, and you can try Vilix free for 7 days to set it up end to end.

Try Vilix free for 7 days

Persistent memory across ChatGPT, Claude, and the AI tools you already use.

Try Vilix free