Lists the user's brains and which one is currently open in the app. Call this first in any session: every other tool needs a brainId and guessing one is not allowed. Also shows whether the semantic index is ready.

A thought in full: note, parents, children, jumps, tags, type and attachments in one call. The main way to understand a thought's context. Call it when you need to know what a thought is about and how it connects, before changing or extending it.

Finds thoughts by meaning. Call it when looking for something by description rather than exact title — for instance to check whether a thought on this topic already exists. Always pass variants: several phrasings of the same concept (synonyms, translation, a broader and a narrower term). They raise recall noticeably, and if the semantic index is not built they are all that works.

Walks the graph several hops out from a thought. Call it when you need to see the whole neighbourhood: which region of the brain adjoins a topic, where new material should slot in, what already sits nearby. Viewing a thought shows one hop only.

The brain's vocabulary: which thought types and tags exist. Call it before creating thoughts so you use this brain's established conventions instead of inventing new ones.

What changed in the brain over a period, in plain language. Call it for digests ("what did I add this week") and to understand what work was in progress before continuing it.

Manages the semantic index: status, build, refresh. Call with action=status when brain_search complains the index is missing; with action=rebuild to build it for the first time or after changing the model; with action=sync to pull in changes (fast, new material only). The first build on a large brain takes minutes.

Creates a thought and sets its note, type and tags in the same call. Call it when adding something new. Always supply parentId — a thought with no links gets lost in the graph. Search first: a suitable thought may already exist, in which case extending it is the better move.

Changes a thought's properties: title, caption, type, colours. Call it to rename or reclassify. It does not touch the note — use brain_set_note or brain_append_note for that.

Replaces a thought's note entirely. The previous text is lost — to add rather than replace, call brain_append_note. Read an existing note with brain_get_thought before overwriting it.

Appends text to the end of a note without erasing anything. The main way to accumulate material on a thought. If no note exists yet, one is created.

Links two thoughts. Call it when you spot a meaningful relationship between existing thoughts — links are what make a brain a graph rather than a list. Label the link via name: "motivates", "solved by", "contradicts".

Attaches and removes a thought's tags. Tags are separate thoughts of a special kind; take their identifiers from brain_list_types_and_tags, and create a new tag via brain_create_thought. A single call can both add and remove.

Attaches a link to a thought. Checks first whether the same URL is already attached and avoids duplicates. Leave name empty to take the title from the page.

Opens a thought in the user's TheBrain app on screen. Call it to show the person a result — for instance a structure you have just created, so they can see it in the graph.

Permanently deletes a thought along with its note and links. There is no undo. Requires human confirmation. Do not call it to tidy up on your own initiative — only when the user explicitly asked for a deletion.

Writes a whole structure of thoughts, links, notes and tags in a single call. Call it once you have decomposed some material (an article, a document, a discussion) into several connected thoughts — creating them one by one at that volume is slow and breaks halfway. Thoughts reference each other by tempId, arbitrary labels you invent yourself; the server substitutes real identifiers. A parent may be either a tempId from this batch or the UUID of an existing thought, which is how a new branch grafts onto what the brain already holds. Search with brain_search first: some of the thoughts may already exist, and linking to them beats creating duplicates.