SheetDB vs sheet2api: two spreadsheet APIs with MCP for AI agents

Short answer: SheetDB and sheet2api are the two closest products in this category. Both turn a spreadsheet into a hosted REST API that returns rows as JSON objects with your column names unchanged, both filter rows with wildcards, negation and numeric comparisons, and both run a remote MCP server so Claude, ChatGPT or Cursor can read and write the sheet after an OAuth sign-in. The differences are in the details. sheet2api works with Google Sheets and Excel Online; SheetDB works with Google Sheets only. SheetDB adds sorting and pagination, batch writes, a read cache with a duration you choose, CORS and IP restrictions, separate agent permissions per API and a server-side approval step before an agent deletes or mass-updates rows, and it has a free plan. sheet2api requires a filter on every agent delete but has no approval step of its own, and its pricing page lists no free plan. Choose sheet2api when your data lives in Excel Online; choose SheetDB when you want the agent's destructive actions to pass through a human, or when you need queries, caching and production controls on a Google Sheet.

At a glance

sheet2api facts are taken from sheet2api's public documentation, MCP documentation, blog and pricing page as of November 2026. Check their site for current details.

Where sheet2api is a good choice

sheet2api is the option when your workbook lives in Excel Online: sign in with Microsoft, paste the link and you get the same kind of JSON endpoint, and the same MCP server, for a Microsoft 365 spreadsheet. SheetDB does not read Excel files. The REST API is familiar if you have used SheetDB: a plain array of objects, filters in the query string with wildcards and comparisons, updates and deletes by filter rather than by row number. Reading single cells by address (/cells/A1,B2) is available both over REST and to agents. Every API comes with auto-generated documentation, and there is a Python client.

Where SheetDB goes further

• Human approval for agent deletes. delete_records deletes nothing by itself. It counts the matching rows and returns a preview written by the server; the agent confirms small deletions in the chat and sends larger ones to the SheetDB panel, where you approve or reject them while signed in. sheet2api's delete tool runs as soon as it is called; the documentation requires a filter and asks the agent to list the rows first, which depends on the model doing what the docs say. Details: the approval model on the MCP page.
• Agent permissions of their own. In SheetDB the first agent connection adds an AI agents section to every API, with agents allowed to read and nothing else until you say so. Those switches are separate from, and never wider than, the API's HTTP permissions, so you can keep POST open for a form on your website and still deny the agent writes. In sheet2api the agent uses the same four toggles as the REST API.
• Queries. ?sort_by=age&sort_order=desc and ?limit=20&offset=40 run on the server (read, search), and the MCP tools take the same sorting and paging arguments, so an assistant can page through a long tab instead of pulling it whole. sheet2api's public docs describe limit but no sorting or offset.
• Batch writes. One POST appends many rows and one batch update changes many rows; JSON import fills a sheet in one go. sheet2api's REST and MCP docs append one row per call.
• Caching you control. Read responses are cached for 15 seconds by default and you choose the duration or purge the cache (cache), which keeps a busy page or a chatty agent away from Google's per-minute quotas. Plan quotas and rate limits are on the limits page.
• Production controls. Lock an API to your domain with CORS origin restriction, allow only some IP addresses, switch off writes with permissions, and protect the endpoint with a Bearer token.
• Ready integrations. A free WordPress plugin, an HTML snippet that renders rows on any page, guides for Bubble, Adalo and n8n and Make, and an OpenAPI specification for agent frameworks that do not speak MCP.
• Free plan. 2 APIs and 500 requests a month without a card, MCP included, so you can connect an assistant to a sheet before deciding anything.

What an AI agent can do in each

Both services expose one MCP server address for every account and sign the user in with OAuth, so there is no API key to paste into Claude, ChatGPT or Cursor. From there the two designs diverge.

The difference that matters most is the step between "the agent decided to delete" and "the rows are gone". Every MCP client has a mode in which a destructive tool runs without asking, so a server that executes on the call is relying on the model to be careful. SheetDB keeps that decision on the server, scales it with the size of the change and, above the threshold, moves it to a signed-in page that the model cannot click. Guides per assistant: Claude, ChatGPT, Cursor; the reasoning behind the whole setup is on Google Sheets for AI agents.

How simple it is with SheetDB: read adults, add a row

The task: a sheet with columns id, name, age, comment. Get the people older than 18, then add one. With SheetDB the response is exactly what is in your sheet:

• No wrapper key. The response is a plain JSON array, as in sheet2api.
• No filtering in your code. Comparisons like age=>18 go into the URL and only matching rows are returned.
• Many rows in one request. data is an array, so one POST appends one row or a thousand.
• One endpoint for the whole spreadsheet. Pick the tab with ?sheet= instead of keeping one URL per tab.

Here is the whole thing, using plain fetch(). It runs in Node.js or in the browser, and the endpoint below is a real demo API you can try right now:

const api = 'https://sheetdb.io/api/v1/58f61be4dda40';

const params = new URLSearchParams({ age: '>18' });
const adults = await fetch(`${api}/search?${params}`).then((response) => response.json());
// [{ "id": "2", "name": "Alex", "age": "24", "comment": "" }, ...]

await fetch(api, {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({ data: [{ id: 'INCREMENT', name: 'Mark', age: '35', comment: '' }] }),
});

That is the entire integration: one GET to /search?age=>18 returns the adults, one POST with a JSON body appends Mark, and INCREMENT fills in the next id for you. The new row shows up in the spreadsheet immediately. In a chat, the same two steps are search_records with age: ">18" and create_records, once you have enabled create for that API in its AI agents section.

Updating and deleting work by any column value: PATCH /api/v1/{id}/id/4 changes the row whose id is 4 and DELETE /api/v1/{id}/id/4 removes it, even after someone sorts the sheet (update, delete). Language guides: Node.js, PHP.

FAQ

[
  {
    "id": "1",
    "name": "Tom"
  }, …
]