instibrain
Sign in
instibrain  /  connect  /  codex

Codex, with the context no model ships with.

Codex can write the release notes, the migration, the reply to the customer. What it cannot know is your voice, your floor, or why the schema looks that way. The brain is where those are written down, and Codex reads them before it drafts.

Add the brain → What it reaches

one command · or two lines of toml · sign in once

style/voice.mdopen area
askedDraft the release notes for 2.4.
codex, reading the pageDrafted in the terms of style/voice.md: present tense, one change per line, the reason beside each, no adjectives. Three lines, from the merged pull requests tagged 2.4.
Who can read itCodex in any terminal you sign in from, and every other agent of a person the style folder is shared with.
the connection

One command in the terminal, or two lines in a file.

Codex keeps its MCP servers in ~/.codex/config.toml. The command writes the entry for you; the panel shows what it writes.

  1. Add the server. codex mcp add instibrain --url https://brain.instibrain.ai/mcp. Nothing else installs.
  2. Sign in once. The first call opens the brain's sign-in in your browser. Approve it as yourself, and Codex holds permission to read.
  3. Start it in the folder. Run codex in ~/instibrain and the pages are also files in its working folder, so a question is answered from the file or from the brain, whichever is nearer.
a terminalthen ~/.codex/config.toml
$ codex mcp add instibrain --url https://brain.instibrain.ai/mcp # what the command writes [mcp_servers.instibrain] url = "https://brain.instibrain.ai/mcp"
what it reaches

What Codex reaches.

The open area over MCP from any terminal, and the files on a computer that holds the key.

Over MCP: the open area

Every page of an open brain shared with you, by search and by path, from any terminal you sign in from.

In the folder: the files

On a computer running the sync client, ~/instibrain holds every page your key opens as markdown, encrypted-area pages included. Codex reads them like any other file.

The brain's rules first

The handshake tells every agent, Codex included, to read CLAUDE.md at the root of the brain before anything else. Your team writes that page; Codex follows it.

Writes under its own name

Approve writing and Codex records what a session settled as a page. The brain stores the credential that wrote it, so the cockpit shows Codex, not an anonymous write.

before you add it

What a Codex user asks before the first command.

Does Codex need Node or the sync client?
Neither, for the open area over MCP. The sync client is needed only where you want encrypted-area pages as files, and its installer brings its own Node.
Which tools does Codex get?
Codex sees the eight reading tools after sign-in. The seven writing tools appear only when you approve writing, so a read-only connection cannot call them. Each is on the tool reference.
Can Codex and Claude Code share the brain?
Yes. Both add the same address, each signs in on its own, and both read the same pages the same minute. Add a third and nothing in the brain changes.

The draft in your voice, the migration under your constraints.

Connect once, and Codex reads the page before it writes the line.

Get started → Every tool and refusal