instibrain
Sign in
instibrain  /  connect  /  cursor

Cursor reads the decision before it edits the code.

The constraint that looks like a mistake is the one somebody put there on purpose. With the brain connected, Cursor reads the decision record before it proposes the cleanup, and it reads your naming rules before it names anything.

Add the brain → What it reaches

three lines · global or per project · sign in once

decisions/naming.mdopen area
askedRename the scope column to area? The docs call it the area everywhere.
cursor, reading the pageLeave it. decisions/naming.md: scope is the wire name and stays in code and in the database; area is the word prose uses. A rename on the wire breaks every client that holds the old name.
Who can read itCursor on any computer you sign in from, and every other agent of a person the decisions folder is shared with.
the connection

Three lines in one file.

Cursor reads its MCP servers from mcp.json: ~/.cursor/mcp.json for every project on this computer, or .cursor/mcp.json inside one project for that project alone. Either way the entry is the same.

  1. Write the entry. Open Cursor Settings, then MCP, and add a server, or put the three lines in the panel into the file yourself.
  2. Sign in once. Cursor opens the brain's sign-in the first time it connects. Approve it as yourself, and Cursor reads the pages you may read.
  3. Ask in the chat. Cursor calls the brain's tools before it answers: a search, then the page the search found.
~/.cursor/mcp.jsonor .cursor/mcp.json in a project
{ "mcpServers": { "instibrain": { "url": "https://brain.instibrain.ai/mcp" } } } // the same address every client takes; Cursor asks you to sign in once
what it reaches

What Cursor reaches, and from where.

Over MCP it reads what our server can read. In the editor it reads what your key opens.

Over MCP: the open area

The pages of every open brain shared with you, by search and by path, from any computer you sign in on. No key is involved.

The folder, open in the editor

Open ~/instibrain in Cursor and every page the sync client wrote is a file in the tree, encrypted-area pages included: the key is on this computer, so the files are plain.

Decisions before edits

A decision record is a page. Cursor reads it when the question touches that part of the code, so the reason outlives the person who knew it.

Writing is approved apart

Ask Cursor to record what a refactor settled, and once you approve writing it lands as a page under Cursor's own credential.

before you add it

The questions a Cursor user asks first.

Global or per project?
Both work. ~/.cursor/mcp.json connects the brain in every project on this computer. .cursor/mcp.json inside a repository connects it there alone, and travels with the repository if you commit it.
Does Cursor need the sync client?
Only for encrypted-area pages, which Cursor reads as files on a computer admitted to the folder's key. The open area it reads over MCP from anywhere.
Which tools does Cursor see?
The eight reading tools at sign-in, and the seven writing tools once you approve writing. A connection that may only read is never shown a write tool, so there is nothing for it to call by mistake. The tool reference has each one.

The reason behind the constraint, in the editor about to remove it.

Connect once, and Cursor reads the record before the refactor.

Get started → Every tool and refusal