A new skills library and MCP upgrade let any AI model design bases, build interfaces, and wire automations in Airtable — not just read and write records.

Airtable's MCP server has always let AI models read and write records. Now it goes further: paired with a new skills library, any model — Claude, ChatGPT, Gemini, or another MCP-compatible tool — can scaffold a workspace, build interfaces, and set up automations from a plain-language description.

Key takeaways

What is Airtable's MCP server

Airtable's MCP (Model Context Protocol) server is Airtable's implementation of the open MCP standard, which lets AI models securely connect to external tools and data. Instead of copying data in and out of Airtable by hand, a connected model can list your bases, read and filter records, and — with this update — create new tables, interfaces, and automations directly. It respects your existing Airtable permissions, so a model can only see or change what you already have access to.

From records tool to builder: What changed

Until recently, MCP's job stopped at records: list bases, read rows, filter, update, insert. The tool list has since grown to include schema and surface-level actions — creating tables and fields, building interface pages, and setting up automations — so a model can go from "here's what I want" to a working system without a person clicking through the UI at every step.

Two things distinguish this from a model just writing to your base blind. First, field creation is schema-aware: point it at linked tables and it can add count, lookup, and rollup fields on its own, correctly identifying which linked-record relationship to use. Second, the bigger surface-level actions — creating interfaces and automations — ship with guardrails: new automations land as drafts that require manual review before going live, and interface pages go through the same publish/delete lifecycle a person would conduct manually.

The skills library: Instant Airtable expertise for any model

Models must learn how your data is structured before being able to act, but a skills library mitigates this step. A couple of foundational skills — airtable-overview and airtable-filters — teach a model the basics of how Airtable data works: bases, tables, fields, how to query them. On top of that, domain skills like product-ops, marketing-ops, and sales-ops bundle in that same foundation plus built-in expertise for a specific kind of work, from launch coordination to campaign tracking. Install a domain skill, and the model already knows how a launch tracker, a content calendar, or a sales pipeline is supposed to be shaped in Airtable.

Case in point: Building a product launch system from a description

Point a model with the marketing-ops skill installed at a workspace and describe what you want — like a fully functional content calendar— and it scaffolds the structure itself: content, campaigns, channels, events, team members, and production tasks, all linked together. It doesn't start from placeholder data either — it takes live input, like your events page and blog and backfills the calendar with upcoming and previously published content, with every record linked back to its source. From there, it builds the interface the team lives in: a dashboard with calendar, kanban, and timeline views, without requiring anyone to configure each component separately in Airtable. The same protocol works with other use cases , like a product launch tracker, a sales CRM, or any other domain skill. Because MCP is an open standard rather than a proprietary integration, any MCP-compatible model — like Claude, ChatGPT, Gemini, or another connected assistant — can use the same server and the same skills library to get the same result.

Prompt packs for every persona

The skills library contains ready-to-use prompt packs — "Ask Claude" and "Ask ChatGPT" versions — so product, marketing, and sales teams each get tested prompts written for their own launch work instead of starting from a blank cursor. Grab the pack for your team, point it at your workspace, and go.

From template to prompt: Dashboards and Slack handoffs on demand

Popular Airtable templates — a launch dashboard, a project tracker — can now be spun up by describing what you want in natural language instead of configuring them field by field.

Getting started with Airtable's MCP server and skills library

Head to the connectors and MCP server setup docs to connect a model, and to the agent docs to install skills. Airtable's new agent-focused documentation home covers all of this — every page is copyable, so you can drop it straight into whatever model you're using.

Build your next system with Airtable

The pattern across all of this is the same: describe the system you want, and let the model build it — tables, interface, automations — inside Airtable, using your permissions and your real data.

Continue to Day 2 of this series: MCP for your whole team — how form submissions and interface reading extend this from builders to everyone else.