Givebutter

Givebutter documents it as read-only in three separate places, stating that connections can never create, edit or delete anything in your account. We did not read the tool annotations ourselves, because the server requires authentication, so this is Givebutter's claim rather than our observation. The connector's one OAuth scope is named mcp, which does not distinguish reading from writing.

Yes. Contacts are one of the nine data types the connector can query, and Givebutter's own example prompts include finding contacts in a state who gave above an amount. Giving history, location and tags are all listed as queryable. Whatever your connected account can see, the assistant can retrieve and place into your AI client's conversation.

No, but you do need to be allowlisted. Givebutter names no plan requirement anywhere in its AI Connections documentation. The gate is the beta instead: the docs state access is limited and your account must be enabled before you can authorize a connection. If sign-in fails with a limited-access message, contact Givebutter support.

Exactly one. Givebutter states each connection is scoped to a single account that you choose during sign-in, and that the assistant can only see data in that account. To query a second account you add a separate connection and pick that account when you sign in. The docs list this as one of three protections on the connection.

Up to 100. Givebutter documents a per-query row cap of 100 and a default of 25, with a limit of 0 returning aggregate totals only. For larger questions Givebutter directs you to ask for a summary or a top-N list so the assistant aggregates rather than listing every record. Paging through results uses an offset.

Roughly 120 list queries per minute and 30 summary queries per five minutes, according to Givebutter's documentation. The docs frame these as responsiveness limits and say to wait briefly and ask again if you hit one. Separately, the unauthenticated endpoint returned an x-ratelimit-limit of 5 to our anonymous probe on 2026-09-08.

A structured query object, not SQL and not free text. Givebutter documents required entity and filters parameters plus optional aggregates, fields, sort, limit and offset. Your assistant builds that object from your plain-language question after calling givebutter_load_entity_context to learn which fields exist and how each one can be filtered, grouped and aggregated.

That is the expected first-request behaviour. Givebutter uses OAuth, so the server refuses any call without a token and your client opens a browser sign-in. Our anonymous handshake on 2026-09-08 returned HTTP 401 with a Bearer challenge, confirming it. Complete the browser sign-in once and the client reconnects automatically.