Google Sheets

Google Sheets for fancy-flow — as four imported, versioned packages, one per runtime. Not vendored source: a copy cannot be upgraded, and third-party APIs change.

The ui package is the editor surface and is React on every host — a PHP or Python project installs it and its own runtime package, and never the js one.

What it costs you

One dependency: @particle-academy/fancy-connector-core (or particle-academy/fancy-connector-core on Composer), which the js and php packages pull in themselves. The Python package has zero runtime dependencies.

No Google Sheets SDK. Plain HTTP, deliberately: a vendor SDK is third-party code subject to the kit's full approval bar, and one per provider is hundreds of dependencies nobody is tracking.

Setting it up

Everything below is generated from provider/manifest.json, so it cannot disagree with what the packages do.

Credentials

A Google Sheets connection holds 4 values.

Two kinds of value, and mixing them up matters. A provider credential is ONE value for the whole installation — an OAuth app's client secret serves every connected account. An account credential is one per connected account. A host that stores the second where it stores the first lets one account's credentials reach another's.

Authorising

Google Sheets uses OAuth2 (authorization_code). The package DECLARES the exchange; the HOST performs it — a consent screen needs a browser, a redirect URI and somewhere to persist the result, and all three belong to the host.

• Authorize URL — https://accounts.google.com/o/oauth2/v2/auth
• Token URL — https://oauth2.googleapis.com/token
• Scopes — https://www.googleapis.com/auth/spreadsheets
• Access token lifetime — 3600 seconds (1 hours). A host that never refreshes works all afternoon and is broken by morning.

The refresh tokens do not rotate: the same one is reusable, so a refresh may safely be retried and may run concurrently. Stated rather than assumed, because the opposite — a provider that spends the token and revokes the grant on a replay — looks identical until it happens.

The estate

Google Sheets has no test estate, and somebody checked. Everything this connector does is real. Use the faker to build against it.

Google has no sandbox for Sheets. A test spreadsheet is a real one in a real Drive, so every append is real -- point this at a scratch sheet, not a production one. The faker is the only way to develop against it without touching a document.

What it can do

Actions

row_append — Google Sheets row

Append a row to a Google Sheet.

POST /v4/spreadsheets/{spreadsheetId}/values/{range}:append · unsafe to replay — a retried durable run does it TWICE

values_get — Read Google Sheets values

Read the cell values of a range.

GET /v4/spreadsheets/{spreadsheetId}/values/{range} · reads only — safe to replay

values_update — Write Google Sheets values

Overwrite the cell values of a range. Existing values in that range are replaced, not pushed down.

PUT /v4/spreadsheets/{spreadsheetId}/values/{range} · idempotent — safe to replay

Run it before you have credentials

Every operation ships a faker, whether or not Google Sheets has a sandbox. Set a node's mode to fake and it returns the shape Google Sheets actually publishes — the same field names, deterministically — so you can wire the downstream nodes before touching an account, a key, or a network.

This repository is generated

provider/ is the source. Everything under packages/ is emitted from it and must not be hand-edited — CI regenerates and diffs on every push, and the next protocol sync destroys anything it finds. See AGENTS.md.

Two namespaces, which do not match on purpose

The repo is github.com/Fancy-Friends/google-sheets; the packages publish under particle-academy. Nothing derives one from the other — the names come from weaver's friends.json and nowhere else.

Licence

MIT.