Simple, elegant and beautiful — like the interfaces from before AI was plastered all over competitors (Airtable, for example). A live grid, typed columns, spreadsheet-compatible formulas: a web database you actually use. All of the AI powers live directly in the Agents you already use: Claude, ChatGPT, OpenClaw, Hermes. MCP is the AI surface, entirely outside the product. No copilot. No AI credits to buy.
Free during beta. No card. MCP in one paste — bips.ai/mcp
| Nametext | Categoryselect | Areatext | Paramsnumber |
|---|---|---|---|
| create_table | Write | table | 4 |
| run_query | Read | query | 3 |
| save_version | Write | history | 2 |
| undo | Destructive | history | 1 |
| migrate_airtable | Write | airtable | 3 |
The grid is a first-class human surface — you open it, you type, formulas compute. No in-product AI chrome to buy. MCP is fully externalized: there is no in-product copilot, and there are no AI credits to purchase. All of the AI powers attach to Claude, ChatGPT, OpenClaw, Hermes and the rest, on the subscription you already pay for.
Typed columns, links, views, forms, history. Spreadsheet-compatible formulas that compute on read. A grid your team edits together, live, with cursors — whether or not an assistant is connected.
The web surface is the product, not a preview of what the assistant just built.
67 tools covering the whole product — tables, columns, views, forms, automations, comments, undo. Your assistant creates the table and fills it, through MCP, not through a sidebar we sold you.
It can undo itself: save_version before it starts, undo when you hate the result.
There is no AI inside Base and nothing extra to meter. No copilot, no sparkle icons, no credit pack on top of Claude, ChatGPT, OpenClaw, Hermes or Grok.
Automations keep firing when you're asleep and your assistant isn't running. People who have never heard of MCP can still fill a form.
Model Context Protocol is the AI surface, entirely outside the product: a standard so Claude, ChatGPT, OpenClaw, Hermes and the rest can use Base as a tool. You paste one URL into the assistant you already pay for. You approve it once. There is no in-product copilot, no sparkle overlay, no API key, and no AI credits to buy.
https://bips.ai/mcpClick to select, or Copy. Same URL for every agent that speaks MCP. There is no API key to create, paste, or leak.
Claude: Settings → Connectors → Add custom connector. ChatGPT: Settings → Connectors → Add. OpenClaw, Hermes, Grok, or Gemini: the same URL.
Your assistant opens AuthKit. You sign in and approve, like any other connector. WorkOS issues the token; Base verifies it. You can revoke it any time.
67 tools covering tables, columns, views, forms, automations, comments, undo. It can save_version before it starts and undo when you hate the result.
Every table is a typed, read-only SQL view. Your assistant writes a SELECT and gets the answer, instead of paging thousands of records through its context to count them.
Read-only is enforced four independent ways. Writes stay on the typed tools, where validation and history live.
There is no AI inside Base. Scoped access, field-level permissions, and a full change log of every write with the actor attached. Everything an assistant does is attributable and reversible.
Branching undo means a bad instruction is a step backwards, not a restore from backup.
It is an Airtable-class database — live cursors, branching undo, time travel, formulas computed on read. A small team; the infrastructure is not.
Base is an Airtable-class database, so the model you know comes with you — tables, fields, links, views, formulas. Paste a read-only token and a base ID and the copy takes about a minute.
Migration reads, never writes, so you can run both while you decide. Seventeen field types map directly with their options intact. We tell you plainly what doesn't come across — computed fields arrive as values, and views need rebuilding.
Base is free during the beta, no card required: unlimited bases, editors, viewers and agents, forms, automations, history, MCP, and the Airtable migration. We are using the beta to gather feedback and usage data before we settle on a pricing model, and every account will hear from us before anything changes. Nothing is gated or counted today: editors, viewers, form submitters and agents all come with it.
A single base over 100,000 rows: email us first. Reads get expensive at that size and we would rather talk it through than find out together. Email us.
Free during the beta, with notice before that changes. Cancel any time; your data leaves by CSV or REST whenever you want it.
You'd find this out in week two anyway. Better now, from us.