Help Center
Tabulous MCP connector
Connect Claude, ChatGPT, Cursor, and other AI assistants directly to your Tabulous workspace.
Overview
The Tabulous MCP server lets any AI assistant work with live Tabulous data. It’s built on the Model Context Protocol (MCP), an open standard for connecting AI tools to external systems.
The server is hosted by Tabulous, so there is nothing to install or self-host.
- Server URL
https://app.tabulous.ai/mcp- Transport
- Streamable HTTP
- Authentication
- OAuth — browser sign-in with your Tabulous account
- Works with
- Claude, Claude Code, Codex, Cursor, ChatGPT, and any MCP-compatible client
Prerequisites
- A Tabulous workspace and an active user account. If your team doesn’t use Tabulous yet, book a working session to get started.
- An AI assistant that supports remote MCP servers with OAuth, such as Claude, Claude Code, Codex, Cursor, or ChatGPT.
Set up the connector
Claude Code
Add the server from your terminal:
claude mcp add --transport http tabulous https://app.tabulous.ai/mcpThen run /mcp inside Claude Code to complete the OAuth sign-in in your browser.
Codex
Add the server from your terminal:
codex mcp add tabulous --url https://app.tabulous.ai/mcp
codex mcp login tabulousThe login command opens your browser to complete the OAuth sign-in.
Cursor
Open Settings → MCP & Integrations and add a new MCP server, or add this to your mcp.json:
{
"mcpServers": {
"tabulous": {
"url": "https://app.tabulous.ai/mcp"
}
}
}Cursor will prompt you to authenticate in your browser the first time the server is used.
Claude, ChatGPT, and other clients
Every client tucks custom connectors into a slightly different settings menu, so the easiest path is to ask your assistant to walk you through it:
“Help me add a custom MCP connector. The server URL is https://app.tabulous.ai/mcp and it authenticates with OAuth.”
Authentication
The Tabulous MCP server uses OAuth with a browser-based sign-in. When you connect, you’ll be redirected to Tabulous to sign in with your existing credentials and authorize the connection — no API keys to create, copy, or rotate.
The connection acts on your behalf: everything the assistant reads or stages is tied to your Tabulous user. To disconnect, remove the connector from your AI client’s settings at any time.
Security and permissions
The MCP server is designed so that an AI assistant can be genuinely useful without ever putting your data at risk:
- Your permissions apply. A connected assistant acts as the authenticated user and can only see what that user can see in Tabulous.
- Reads are direct. Browsing tables, records, files, and users happens immediately, so questions get answered fast.
- Every write requires human approval. Assistants never modify your workspace directly. Instead, each change is staged as a changeset — a complete, ordered plan of operations that Tabulous validates and previews in one pass. The changeset lands in Pending Review with a link where a person reviews a row-by-row preview and approves or rejects it in the Tabulous app. Nothing changes in your workspace until a human approves.
- Agents can never self-approve. Approval only happens in the Tabulous app by a signed-in user; agent credentials cannot approve a changeset.
- Invalid plans are rejected outright. A plan that doesn’t match your schema fails validation with per-operation errors, and nothing persists.
For security practices and compliance documentation, visit our trust center.
What you can do
Reading your workspace
These tools read data directly and never modify anything.
| Tool | What it does |
|---|---|
registry_tables_list | List every Registry Table in the workspace with its full schema. |
records_list | Read records from a Registry Table with filters, search, and sorting. |
records_suggest_names | Suggest valid, unused names for new records. |
files_list | Browse the files stored in a record’s folder. |
toolkit_list | List Toolkit Apps, Scripts, and Workflows. |
users_identify | Show which Tabulous user the connection is acting as. |
users_list | List the members of the workspace. |
workspace_info | Show basic workspace details. |
docs | Fetch server-hosted reference documentation for agents. |
changesets_list | List staged changesets and their review status. |
changesets_show | Show the operations and status of a specific changeset. |
Staging changes
All writes — creating or updating records, changing table schemas, editing Toolkit items, or managing files — go through a single staging tool and wait for your approval.
| Tool | What it does |
|---|---|
changesets_stage | Validate a complete plan of record, schema, Toolkit, or file operations and stage it as a pending changeset for human review. Nothing is applied until a person approves it in Tabulous. |
changesets_discard | Withdraw a pending changeset before it is reviewed. |
Utilities
| Tool | What it does |
|---|---|
cli_connect | Install and connect the Tabulous CLI in the assistant’s environment, enabling file downloads and uploads with the same permissions. |
Frequently asked questions
Which AI assistants work with the Tabulous MCP server?
Any AI assistant that supports remote MCP servers with OAuth, including Claude (web and desktop), Claude Code, Codex, Cursor, and ChatGPT. Check your assistant’s documentation to confirm MCP support.
Can an AI assistant change my data without my knowledge?
No. Assistants can read data directly, but every proposed change is staged as a changeset that a person must review and approve in the Tabulous app. Until it’s approved, nothing in your workspace changes — and you can discard a pending changeset at any time.
What data can a connected assistant see?
Exactly what the signed-in user can see. The connection authenticates as you and inherits your Tabulous permissions; it has no elevated or separate access.
Do I need an API key?
No. The MCP server uses OAuth with a browser sign-in. There are no keys to generate or store, and you can revoke access by removing the connector from your AI client.
What can the connector do on its own?
Almost everything: reading tables, records, users, and file listings, and staging any change — records, schemas, Toolkit items, and file moves or deletions — for your review. The one thing it can’t do is transfer file contents: downloading a file to work on it, or uploading a new one, requires the Tabulous CLI.
What is the Tabulous CLI, and when do I need it?
The CLI (command-line interface) is a small program an assistant can install and run in its own environment to do everything the connector does, plus download and upload actual files. An assistant with terminal access — like Claude Code, Codex, or Cursor — can set it up by itself using the connector’s cli_connect tool; just ask it to.
Note that the CLI needs network access to app.tabulous.ai from wherever the assistant runs. Sandboxed environments such as Claude Cowork or ChatGPT Work restrict outbound network access by default, so file uploads and downloads may not work there unless your administrator allows access to that domain.
Support
Questions about connecting an assistant to Tabulous? Contact us — we’re happy to help.