Connections

A connection is anything that can act as you: an agent, an app or the CLI. Every one of yours is on the Connections page (the account menu at the top right, then Connections; its address is /connections), where you can revoke it. The page used to be at /tokens: old links and bookmarks still land on it.

The three types

The Type column on the Connections page shows which one each is.

	App	Token	Reliquary CLI
For	MCP clients that sign in: Claude Code, Claude.ai, ChatGPT	MCP clients that can only send a header: Cursor, VS Code, scripts, headless agents	the `reliquary` CLI on one computer
Made by	signing in from the client and choosing Allow	Create read-only token on the Cursor, VS Code or Other clients tab of Connect, or New token on the Connections page	`reliquary login` and choosing Allow
Reaches	the MCP endpoint	the MCP endpoint	the env API only: variable values, never files
Vaults	all yours, or the ones you tick	all yours, or the ones you tick	all yours, or the ones you tick
Access	read only, or read and write	read only, or read and write	reads values; optionally sends a `.env` for approval
Lifetime	1-hour access tokens, refreshed by the client; at most a year	7 days to a year, fixed when made	1-hour access tokens, refreshed by the CLI; at most a year
Named	the client's name, "from" its site	the name you gave it	"Reliquary CLI"

Apps and the CLI sign in through the web app at https://app.reliquary.redmage.cc, the issuer your client checks. The CLI uses that server unless you name another with --server.

Every connection acts as you, minus the ceiling. It can never do more than your role allows in a vault.

When a client signs in, there is no secret to copy: it signs in through your browser and keeps its own refreshed tokens. Make a token only for clients that can't sign in, and keep it out of chats and config files an agent can read: read it from an environment variable or a password prompt. See Connect Cursor and Connect VS Code.

The Connect page has a tab for each client with only that client's steps. Under the steps, once for every tab, it says what an agent can't do (approve, change rules, manage members or read variable values: the ceiling) and how to keep a token safe.

Make a token

Quickest, for Cursor, VS Code or another client that can't sign in:

On the Connect page, open that client's tab.
Name the token after the agent and machine, then choose Create read-only token or Create read and write token. It reaches all your vaults and expires in 90 days. If you belong to more than one vault, the tab shows Choose vaults and create a token instead, which opens the full form, where nothing is chosen for you.
The next page shows the token once, with the steps for that client and the token already in them.

For some vaults, or another expiry, choose use the full form on the tab, or New token on the Connections page: name it, choose its vaults (if you belong to more than one vault, nothing is chosen for you), Read only or Read and write, and when it expires, then Create token. Copy the token, then choose Done. It is shown once.

Check that it connected

The Connect page says Connected and names a connection used in the last 15 minutes, with when and from which client. If none was, it says so. Reload it after the last step. A revoked or expired connection never counts.

A vault's Overview shows a Connect an agent step to its writers, for the first 14 days after the vault is made, until you have any connection at all (a token, an app or the CLI).

Every vault's Overview also has a Connect an agent button next to New file, so you can connect another agent from any vault at any time. Viewers see it too. It stays there while the step above is showing, too.

If Create token is refused (no name, Only the vaults I tick with none ticked, an expiry past a year), the form comes back with the reason and a reference, and everything you chose is still there to correct. Nothing was created.

Scope is fixed

A connection's vaults and access are fixed when it is made. To change them, revoke it and make another. "All my vaults" includes vaults you join later; ticking vaults limits it to those.

If you belong to more than one vault, the consent page and the New token page don't preselect either choice, and refuse an answer that names neither. If you belong to one vault, All my vaults is already chosen. To keep clients apart, see Keep client work separate.

Revoking

On the Connections page, choose Revoke next to it. A confirm page shows its type, vaults, access and last use; choose Revoke with its name to confirm. It is refused on its next request.

Only you can revoke your own, in person: no agent, token, app or CLI can revoke one (an app can still end its own connection). A vault owner can also cut a member's connection off from their vault on Members, without touching the member's other vaults.

When you leave a vault, or are removed, connections that reached only that vault are revoked, and the vault drops out of connections that reached several.

If one leaked, see Rotate a leaked token.

What the Connections page shows

Live connections come first. Each shows its name, type, vaults, access, when it was last used and by which client (the name the client reports), and when it expires. Expired and revoked ones are folded under Expired and revoked, with when each ended, so you can still see what had access.

Tokens are stored only as hashes: a new token is shown once, when you create it, and never again.