instibrain
Sign in
instibrain  /  connect  /  claude code

Claude Code, with your company's memory.

Claude Code already reads your repository. Add the brain, and it reads the decisions behind the code too: the floor, the runbook, the reason the schema looks that way. One command, a sign-in in the browser, and every session starts from your team's rules.

Add the brain → What it reaches

one command · sign in once · CLAUDE.md first

runbooks/deploy.mdopen area
askedWhy does api restart before web in the deploy script? Can I run them in parallel?
claude code, reading the pageKeep the order. runbooks/deploy.md: api takes the migration lock on start, and web answers requests only once api reports healthy. In parallel, web serves the old schema for a few seconds.
Who can read itClaude Code in any terminal you sign in from: an open-area page reaches every agent of a person this folder is shared with.
the connection

One command, then a sign-in you do once.

Claude Code speaks MCP over Streamable HTTP, and so does the brain. The command registers the brain for your user; a .mcp.json in a project does the same for everyone who opens that project.

  1. Run the command. claude mcp add --transport http instibrain https://brain.instibrain.ai/mcp. That is the whole install: no package, no plugin.
  2. Approve it in the browser. The first call opens the brain's sign-in. Sign in as yourself, and Claude Code reads what you may read, no more.
  3. Ask. The handshake tells it to read CLAUDE.md at the root of the brain before anything else: your team's rules for agents. Then it searches, reads, and answers from the pages.
a terminalthen .mcp.json, for a project
$ claude mcp add --transport http instibrain https://brain.instibrain.ai/mcp Added HTTP MCP server instibrain # or, in a project, for everyone who opens it { "mcpServers": { "instibrain": { "type": "http", "url": "https://brain.instibrain.ai/mcp" } } } # mid-answer, in a session instibrain:search_brain(query: "deploy order") instibrain:read_node(path: "runbooks/deploy.md")
what it reaches

Over MCP, the open area. Beside the sync client, the files too.

Two doors into the same brain, and the key decides which one opens.

Over MCP: the open area

Every page of an open brain you may read, searched and read by path. No key takes part, so this works from any computer you sign in on.

Beside the sync client: the files

Start Claude Code in ~/instibrain and the pages are markdown files in its working folder, encrypted-area pages included, because the sync client on that computer holds the key.

CLAUDE.md, both ways

Over MCP the brain names CLAUDE.md as the first thing to read. In the folder, ~/instibrain/CLAUDE.md imports the brain's own, so each session loads your rules before the first prompt.

Writing is a second yes

Reading comes with the sign-in. Writing back is approved on its own, and every write is recorded under Claude Code's own credential.

before you add it

Three questions first.

Does Claude Code need the sync client?
Not for the open area: over MCP it reads any open brain you may read, from any computer. The sync client matters only for encrypted-area pages, which exist as files on a computer admitted to the folder's key.
Which tools does Claude Code get?
Eight for reading: search_brain, read_node, list_nodes, node_history, list_brains, brain_map, read_skeleton and check_structure. Seven more appear once you approve writing. The tool reference describes each one.
Can the whole team's Claude Code share one brain?
Yes. Each person adds the same address and signs in as themselves, so each one reads the folders shared with them and nothing else, and the cockpit shows which agent read or wrote which page.

Your repository, and the reasons behind it.

The brain is the part Claude Code did not have: what your team decided, and why.

Get started → Every tool and refusal