StackJack 0.1.0.5302+6d74472

This release adds support for holding several named connections of the same connector within one organization, so an MSP that serves multiple customers on one tool can connect each of them and have AI agents work against a specific one per call. Organizations with a single connection per connector see no change in behavior.

New features

• Several connections per connector. On the Connectors page, a connector's drawer now lists every connection you hold for it, each showing its name, default and health badges, server address, and tool calls this billing cycle. From the drawer you can add a connection, edit it, rename it, set it as the default, test it, or remove it. Removing one of several connections keeps the connector connected, and the confirmation tells you which connection becomes the new default.
• Per-call connection selection for AI agents. When a connector has two or more connections, its tools accept a connection argument (by name or key). Calls that omit it go to the default connection. Each tool result includes metadata naming the connection that served it, and a new stackjack_list_connections platform tool lets an agent discover the connections available to it. Errors for an unknown, disabled, not-permitted, or temporarily unavailable connection are documented on the tool errors page.
• Endpoint pins. You can pin an MCP client to one specific connection. A pinned client always uses that connection, is not offered the connection argument, and is refused if it names any other. Pins are preserved when you save unrelated settings, and a pin to a connection that has since been disabled or removed fails closed rather than silently falling back to the default.
• Personal sign-ins and agent enrollment per connection. For connectors that use per-user sign-in, team members see one card per connection and sign in to each individually; sign-ins started from a named connection are tied to that connection. When enrolling an agent's credentials on a connector with more than one connection, you can choose which connection to enroll against.

Improvements

• Connections are fully isolated from each other: cached access tokens, connection-health tracking and auto-disable, duplicate-action protection for destructive tools, and connection tests are all tracked per connection, so one connection's failures or credentials never affect a sibling connection.
• If connection information is temporarily unreadable, a call that names a connection now returns a retryable error rather than a definitive refusal or a result from the wrong connection.
• Entering an invalid connection name or key now shows a plain-English message explaining what a valid value looks like, instead of a generic error.
• Connection names are validated before credentials are saved. If a name cannot be applied after the credentials are stored, the form tells you the credentials saved and the name did not, rather than prompting you to add the connection again.
• Opening one connection's edit form and then another's before the first finished loading no longer shows, or saves, the first connection's values on the second.
• Adding a connection no longer fails or stores an incomplete credential when the connector's default connection is already connected.
• The connector setup guide gains a note about shared connections, and the "How connectors work" and "How tools work" help pages, along with 20 connector pages, now describe multiple connections and the connection argument.