Is MCP the same as a plugin?

MCP is the protocol; a plugin is the packaged connector. Some connectors are MCP-native — the template's deps list names whichever applies.

Do connectors need a paid plan?

Yes — every template with a non-empty plugins or mcp field carries the `pro plan` chip on this site.

The narrowest one that works: drafts-only for drafting agents, read-only for digests. Never grant send scope if a human reviews output.

Pick the agent, open its connectors, authorize the tool (Sheets, Zendesk, Gmail…), grant the narrowest scope that works.

Open the agent that needs the connector — connectors attach per-agent, not per-account.
Find the connector your template names (the deps block lists it under plugins or mcp).
Authorize the account and pick a scope. For drafting agents, choose the scope that creates drafts — not the one that sends.
Point the agent at a specific resource — a named sheet, a queue, a label — not “everything.”
Run once and check what it actually touched before leaving it scheduled.

Draft vs send — invoice chasers, candidate follow-ups, outbound: draft scope only. You click send.
Read vs write — digest agents need read access to the source, not write. A read-only connection can’t damage anything.
One resource vs the account — connect the AR sheet, not “all spreadsheets.” Smaller surface, smaller blast radius.

Connector authorized but agent sees nothing — check that you attached it to the right agent, not just the account.
Output ignores the sheet — the prompt still needs the resource name in its config placeholder.
Runs worked manually, fails on schedule — scheduled runs act under whatever scope the agent holds — re-check what you granted.

Connector attached? Routines and scheduled runs covers the other half of the paid tier.