Frequently asked questions

Nothing. InnerSheet is free and open source under the MIT licence, and the Google Sheets API is free within its quota. The only cost is whatever your AI client charges — there is no per-call billing and no vendor in the middle.

Not with a local install: ChatGPT on the web only connects to remote MCP servers over HTTPS. Deploy InnerSheet in HTTP mode (API-key protected) and add it as a custom connector in Developer mode, which needs a Pro plan or higher. Codex, OpenAI's local CLI, works with the local server directly.

Yes, the same way: deploy in HTTP mode and add it under Settings → Connectors → Add custom connector. The Free plan allows one custom connector. Claude Desktop and Claude Code work with the local server.

Yes — via the raw escape hatch. Google's Sheets API supports creating pivot tables with a batchUpdate + updateCells request carrying a PivotTable definition, and sheets_batch_update passes that request verbatim (Google's pivot tables guide). There is no dedicated pivot tool yet because a correct definition (source range, row/column groups, values) has to be built by hand. Otherwise compute the summary with ordinary reads and writes, or build the pivot in the Sheets UI or Apps Script.

The server runs on your machine and talks directly to Google with your credentials — no vendor sees your data. Safety is layered: dry-run previews, a read-only mode, a spreadsheet allowlist, confirmation for destructive calls, and a hook that blocks deletes. Read the full threat model on the security page, including prompt injection through sheet contents.

Honestly, the Google Cloud OAuth client — creating the project and enabling the APIs takes about five minutes the first time, and it's the step non-technical users stumble on. The one-click installers handle everything else. The install guide walks through the Cloud setup click by click.

A managed platform is the fastest way to start: single-operation tools, per-use billing, and your data flows through their infrastructure. InnerSheet is self-hosted and free, with workflow-level tools that finish a whole job in one call, custom rules like verbatim report copying, and a direct connection to Google.

Any MCP-compatible client that can launch a local process: Claude Desktop and Claude Code, Cursor, Codex, VS Code with GitHub Copilot, Gemini CLI, Windsurf, Cline, Zed and JetBrains IDEs. Web chats — ChatGPT, Claude.ai, the Gemini web app and Muse — need the remote HTTP deployment instead, because they run in the cloud and cannot start a program on your computer. The full compatibility matrix, with copy-paste configuration for each client, is on the AI Clients page.

macOS, Windows and Linux. The install guide is written macOS-first, then Windows, then Linux. You need Python 3.10 or newer, or just install uv with one command and let the installer handle the rest.

Yes. Create a service account in Google Cloud, download its key, share each sheet with the service account's email as Editor, and set SHEETS_MCP_AUTH=service_account with GOOGLE_SERVICE_ACCOUNT_JSON pointing at the key. This is the better choice for servers and shared team sheets.

Writes return exactly what changed (updatedRange and updatedCells), so the mistake is visible in the transcript. For a bad write, fix it with another write or restore from Sheets' version history (File → Version history). For a deleted sheet there is no undo — which is why deletes need explicit confirmation. Using dry_run first prevents most mistakes entirely.

No — Google Sheets is a cloud service, so the server needs internet access to reach Google's API. The server itself runs locally, but every tool call is a network call to Google.

Download the latest zip and replace the folder, keeping your client_secret.json and the cached token. Your AI client's configuration points at the folder, so it picks up the new version on the next restart. Check the release notes for breaking changes to tool names.

Yes — the MIT licence allows it. Note that the MCP Registry only lists metadata (it doesn't host code), and if you run one shared OAuth client for many users, Google's sensitive-scope verification and its 100-user cap for unverified apps will apply. Self-hosted, per-user setups avoid that entirely.