cobrain
Create your brainGo to your brain
Guides · reference

The tools your AI uses

An AI connected to Cobrain has twenty-three tools: fifteen to read, search and be guided, seven to write and reorganise, one to delete. All of them work with your permissions: what you cannot do, it cannot do either.

The connector

A standard remote MCP server, with OAuth 2.1. One address, for every AI.

https://cobrain.space/mcp

MCP over HTTP (Streamable HTTP). The program registers itself, the user authorises on a Cobrain page that states what the program will be able to do, and it receives a grant that expires and renews. No API key. Every tool runs with the identity of the user who authorised it: the same rules as the site, applied by the database.

For a program with no guide of its own

Paste this message into the conversation: it carries the server name, the address and the kind of authentication. Many programs configure themselves; the others tell you where to add it.

Add the remote MCP server "cobrain" to your environment:
https://cobrain.space/mcp

Use OAuth authentication in the browser. If you cannot edit the configuration yourself, walk me through it step by step using the exact menu names.

If the program only accepts a command, the bridge is npx -y mcp-remote https://cobrain.space/mcp. If it asks for an API key, there is none: ask it for a remote MCP server over HTTP with OAuth.

The twenty-three tools

Name, what it does, and whether it reads, writes or deletes. Errors reach the AI in English, suggesting the right path when it gets one wrong: your AI relays them to you in your own language.

list_workspaces

Your brain and the ones shared with you, with the access you have on each. It also says which account the AI is connected as.

reads
list_files

The list of notes under a folder. If the folder does not exist, it suggests similar names.

reads
find_notes

Finds notes and folders by name or by part of a path, like a computer's «find». It also resolves a [[wikilink]].

reads
fetch

Opens a note from the id a search result carried and gives its whole text. It only exists when the connected AI is ChatGPT: it is the half it needs to cite notes with a clickable link.

reads
settings_read

Used by the plugin page in ChatGPT: reads Cobrain's settings to show them. Today that is the Dream's cadence.

reads
settings_update

Used by the plugin page in ChatGPT: saves the setting you changed (every week, every month or never for the Dream). It only touches your brain's settings, never the notes.

writes
read_file

Reads a note, or an attachment: it sees images, reads PDF, Word, PowerPoint, Pages, Keynote, RTF and OpenDocument files as text and Excel sheets as tables, looks at a scanned page as an image, and gets a few frames of a video (not the audio).

reads
load_context

Everything under a folder (or carrying a tag) in one go, instructions first: brain rules, the unit's index card, lessons, the latest decisions, then the notes. It states when it had to truncate.

reads
start_session

The briefing for a unit (a client, a project…): rules, index card, instructions, lessons, latest decisions, open sub-units and the last fifteen changes.

reads
list_units

The units in the brain: path, type, status, how many notes, last change.

reads
recent_changes

Recently changed notes, by whom and when.

reads
list_tags

The tags in the brain, with how many notes each has.

reads
files_by_tag

The notes carrying a given tag.

reads
write_file

Creates a note, or replaces all of its content. It refuses to replace a long note with a much shorter text without confirmation: that guards against an addition mistakenly written as a rewrite.

writes
append_file

Adds text to the end of a note (creating it if missing). This is what the AI should use to add something, rather than rewriting.

writes
patch_file

Changes one part of a note without rewriting it: replaces an exact text with another. If the text is not there, or appears more than once, it refuses instead of guessing. It never creates new notes.

writes
create_unit

Creates a unit (a client, a project, whatever you follow) from a template: the index card and the decision log, born already written (what it is, why you open it, the first decision). On an existing folder it only adds the files that are missing.

writes
attach_file

Attaches a real file (image, PDF, audio) next to a note: it lands in that branch's «allegati» folder and the note gets a reference to it. The file comes as base64, from a public URL, or uploaded separately with a signed link.

writes
move_file

Renames or moves a note, or a whole folder.

writes
delete_file

Moves a note, an attachment or a folder with everything inside it to the brain's trash. The brain's owner can restore it from the trash for 30 days, with its versions; then it is gone for good. The AI cannot see or restore the trash.

deletes
start_project

The script for a guided brainstorm to set up a new unit with you: what to follow, what to call it, how to structure it. It changes nothing on its own.

guides
dream_tidy_up

The Dream: among the notes touched in the last days (1 to 15) or in one folder, finds what to tidy up — notes that say the same thing, decision logs grown too long, links that lead nowhere — and gives the AI the script to fix it with you: it does the obvious things and asks about the rest.

guides

The ones ChatGPT's interface uses

The AI does not use them: ChatGPT calls them when you type @ in the message box or open Cobrain from the sidebar. They only read, with your permissions.

search_mentions

Used by ChatGPT's message box, not by the AI: when you type @ it suggests notes by name (the latest ones you touched if you have not typed anything yet), and the note you pick reaches the AI to read.

reads
open_brain

Opens Cobrain full screen from ChatGPT's sidebar: the brain's tree, search and the open note, with the «Attach to chat» button.

reads
brain_panel

The same Cobrain app, in a tab beside the conversation.

reads
app_note

Gives the Cobrain app the note to show, already laid out: HTML written inside a note comes out as text, never as code.

reads

What they cannot do

For anyone assessing the connector before authorising it.

  • Step outside the user’s permissions. An unshared branch cannot be reached: the rule lives in the database, not in the connector.
  • Erase the history. Every write keeps the note’s previous version; deleting a note leaves its versions behind.
  • Touch the account. No tools to change email, plan, shares or authorisations: that happens only on the site.
  • Talk to other services. The connector reads and writes in the brain and nothing else: no outbound calls, nothing sent elsewhere.
  • Stay connected without permission. The authorisation is revoked from the Connections page, and from that moment every call is refused.

Every call is logged: which tool, with what outcome, in how long. And every note says which program changed it. For the rest, the privacy policy.

Free, no card needed

Connect your AI.

The guides walk you through Claude, ChatGPT, Claude Code, Codex, Cursor and Antigravity step by step. For everything else, the message above.

Create your brain, freeGo to your brainA thousand notes, as many AIs as you want to connect.