Airtable MCP

Any agent. All your business context. One place to work.

Connect Claude, ChatGPT, Gemini, and 3rd party model to Airtable. Work with live data, build apps, and create workflows in the same system your team uses.

  • ChatGPT
  • Cursor
  • Gemini
  • Airtable
  • logo
  • Claude
Capabilities

Ask about the data. Or create the system that runs it

Your agent connects to every part of Airtable: records, structure, interfaces, automations. Find out what's blocked this week, or have it stand up the tables and workflows behind the work, inside the base your team already works in.

  • Work with your data

    Find the right bases, tables, and records. Ask about status, dependencies, and gaps. Then write the answer back into the base.
    Work with your data
  • Build your app

    Create bases, tables, and fields. Build interface pages and record views.
    Build your app
  • Triage feedback into fixes

    Pull feedback from Airtable into your coding agent, ship the fix, and mark the record resolved.
    Triage feedback into fixes
Governance

Works within your permissions

If you can't edit a base, neither can your agent. Anything it builds starts paused until you say go.

  • Uses Airtable permissions

    Uses Airtable permissions

    A connected assistant can only perform actions allowed by your Airtable access and the authorization you grant.
  • Review before workflows run

    Review before workflows run

    Automations created through MCP arrive paused. You approve them and switch them on in Airtable.
  • Managed centrally

    Managed centrally

    Organization admins can block MCP access using the available admin controls.

You have questions? We have answers

A hosted service at mcp.airtable.com/mcp that lets any MCP-compatible AI client read and act on live Airtable data and structure. Your agent can query records, build tables and fields, create and publish interfaces, and write automations.

Nothing. Today, MCP is available on every plan, including the free plan, at no additional cost.

It works with any AI via MCP servers, with out of the box connections for Claude, ChatGPT, and Gemini.

No. Major clients connect over OAuth with a single sign-in. Nothing to generate, paste, or rotate. You can also create granular Personal Access Tokens to further scope access.

Exactly the permissions of the person who connected it, and nothing more. MCP is not a separate access model. If you cannot edit a base, neither can your agent.

It can build. Agents create bases, tables, and fields, build and publish interface pages and forms, and write automations, alongside reading and updating records.

It arrives paused. A person reviews it in Airtable and activates it before it ever runs.

Yes. Admins can block MCP access across the organization or an org unit, the same way they block API access.

MCP runs through the same public REST API, so the standard limits apply: 5 requests per second per base and 50 requests per second per token. Records an agent creates count against your base's normal record limits.