# Google Sheets and MCP from Python Two very different jobs get called "Google Sheets MCP in Python": building a server, and calling one. This covers both, and is honest about when writing your own is a waste of an afternoon. *Last updated: 2026-07-13 · Source: * ## Key facts - Google's API requires a **Google Cloud project plus OAuth credentials or a service account** before it will return a single row. ([source](https://developers.google.com/workspace/sheets/api/guides/authorizing)) - The Google Sheets API allows **300 read requests per minute per project** and **60 per minute per user**. Past that it returns `429 RESOURCE_EXHAUSTED`. ([source](https://developers.google.com/workspace/sheets/api/limits)) - Google's Sheets API is **cell-oriented**: it reads ranges like `A1:D50`, not records. Reassembling those into row objects is work you do yourself. ([source](https://developers.google.com/workspace/sheets/api/limits)) - The MCP half is the easy half. The expensive half is Google — credentials, quota, and the cache you end up writing because **agents are chatty**: one question becomes a schema read plus several queries. ## Do you want to build a server, or use one? If you want to **build** a Google Sheets MCP server in Python, the MCP Python SDK plus `gspread` will get you there — you define tools, authenticate to Google, and run the process. Do this when you need custom tools or write access. If you want to **use** one, you do not need Python at all: MCP clients take a URL. But Python is often how you consume the same data outside the agent, and PasteSheet serves the identical rows over plain HTTP. ## Reading the same sheet from Python The endpoint your agent queries over MCP is also a REST API. No client library, no credentials: ```bash import requests rows = requests.get( "https://pastesheet.com/api/your-endpoint-id", params={"status": "shipped", "sort": "-date", "limit": 50}, ).json() for row in rows["data"]: print(row["customer"], row["total"]) ``` ## What building it yourself actually costs The MCP part is the easy half — the SDK is pleasant and a few tools is not much code. The expensive half is Google: a Cloud project, the Sheets API enabled, a service account and its JSON key, and then quota management. That quota is the part people underestimate. Google allows **300 read requests per minute per project and 60 per minute per user**, then returns a `429` ([published limits](https://developers.google.com/workspace/sheets/api/limits)). Agents are chatty — one question becomes a schema read plus several queries — so a naive server hits this quickly, and now you are writing a cache too. ## What it costs **MCP is included on the Free plan.** Connect any **public** endpoint over MCP with no Google Cloud project and no credit card. Private endpoints and the account-wide workspace server need a paid plan (from **Starter ($9/mo)**) for the keys and OAuth they authenticate with. Pro adds full-text search and aggregation (`count`, `sum`, `avg`, `group_by`) that your AI agent can call through `query_rows`. **Free — $0/month.** For side projects and trying things out. - Endpoints: 3 - Requests: 2,000 / month - Rows per endpoint: 500 - Tabs per endpoint: 1 - Rate limit: 60 / minute ## Frequently asked questions ### Can I write my own Google Sheets MCP server in Python? Yes. The MCP Python SDK plus a library like gspread is a reasonable stack. Expect the Google side — Cloud project, service account, API quota, caching — to take more of your time than the MCP side. ### Do I need Python to use PasteSheet with an AI agent? No. MCP clients connect to a URL, so no code is involved. Python only enters the picture if you also want to read the same rows from a script, which you can do with a plain HTTP request. ### What is gspread? A popular Python library for the Google Sheets API. It is a good choice if you are talking to Google directly, and it still requires a Google Cloud project with credentials. ### How do I avoid Google's API rate limits? Cache. Google allows 300 reads per minute per project and 60 per user, then returns 429. Reading the sheet once and serving repeat queries from a cache removes the problem — which is what PasteSheet does for you. ## Sources - [Authorize requests — Google Sheets API](https://developers.google.com/workspace/sheets/api/guides/authorizing) — Google - [Usage limits — Google Sheets API](https://developers.google.com/workspace/sheets/api/limits) — Google ## Related guides - [How to Read a Google Sheet in Python](https://pastesheet.com/guides/read-google-sheet-in-python) — Three ways to read a Google Sheet from Python: gspread, the official API, and a plain HTTP endpoint. Working code and the gotchas for each approach. - [Self-Hosted Google Sheets MCP Alternative](https://pastesheet.com/guides/self-hosted-google-sheets-mcp-alternative) — A hosted alternative to self-hosted Google Sheets MCP servers: no service account, Docker, or uvx — paste a share URL and get a read-only endpoint. - [Google Sheets API Rate Limits (60/min)](https://pastesheet.com/guides/google-sheets-api-rate-limits) — Google Sheets API rate limits are 300 reads per minute per project and 60 per user. Here is why read-heavy apps hit a 429 — and how caching removes the wall. - [Google Sheets MCP Server for Claude & ChatGPT](https://pastesheet.com/guides/google-sheets-mcp) — Turn any public or private Google Sheet into an MCP server so Claude, Cursor, and ChatGPT can read and query it in plain English. No code, no backend. --- [PasteSheet](https://pastesheet.com) turns any Google Sheet into a live REST API and MCP server for AI agents — no backend, no code. Canonical HTML version of this page: