Connect your tools | Hivemind docs

Terminal

npx @get-hivemind/cli init

It opens a browser, signs you in, detects what is installed, and writes each tool's configuration in whatever format that tool expects. Restart the tools afterwards so they pick up the change.

What it detects

Claude Code · Claude Desktop · Cursor · VS Code · Windsurf · Zed · Cline · Roo Code · Codex CLI · Gemini CLI · Amp · OpenCode · Witsy

Anything not installed is skipped and reported, so you can see what it looked for.

Commands

`init`	Sign in and configure everything detected
`init --dry-run`	Show what would change, write nothing
`status`	Which tools are connected, and to which account
`sync`	Write this project's memories into a block of `AGENTS.md`
`pause`	Stop Claude Code capturing and recalling, everywhere or with `--here`
`resume`	Turn it back on
`signout`	Remove Hivemind from every tool it configured

Flag
`--with google\|github`	Skip the provider chooser
`--key hm_live_...`	Use an existing key instead of signing in
`--only cursor,vscode`	Restrict to specific tools

On a machine with no browser

init signs you in through a browser on the machine it runs on. A server, a CI job or an agent in a sandbox has none. In the dashboard, open Tools, choose Add a tool, and create a key under A machine with no browser. It gives you the command to run there:

Terminal

HIVEMIND_API_KEY=hm_live_... npx @get-hivemind/cli init

Run without a key where nobody is at the terminal, init stops and says this. It exits 3, which means a person is needed and trying again will not help.

What it does to your files

Configurations are merged, never replaced, your other MCP servers and editor settings are left alone. The first time a file is touched, a backup is written next to it as <config>.hivemind-bak.

Where a tool owns its configuration while running and rewrites it as it goes, Claude Code does this, Hivemind goes through that tool's own CLI rather than editing the file underneath it.

Checking it worked

Terminal

npx @get-hivemind/cli status

That reads each tool's configuration. To test that the whole path works, the server, your key, each tool and the Claude Code hooks:

Terminal

npx @get-hivemind/cli doctor

Every line that is not passing comes with the command that fixes it. Add --json to get the same report for a script or an agent to read. It exits 1 when something is failing.

status also shows which account each tool is using. If one is on a different key it says so, in red, that is the failure that otherwise looks like the product is broken, because a tool on the wrong credential sees none of your memories and gives no clue why.

Adding another machine

Run init again on the new machine and sign in with the same account. Each machine gets its own credential; all of them reach the same memory. You can see and revoke them under Connected tools in the dashboard.

Pausing it

Some days, or some repositories, you do not want anything read or saved:

Terminal

npx @get-hivemind/cli pause --here    # this project only
npx @get-hivemind/cli pause           # everywhere
npx @get-hivemind/cli resume          # back on

It stops the Claude Code hooks at once, on your machine, with no network call. status and doctor tell you while it is on, so a paused install is never mistaken for a broken one. Tools that call Hivemind themselves, and the browser extension, have their own switches.