Airtable hosts its own MCP server, available on all plans. It covers records, bases, interface pages and automations, and it respects the permissions the signed-in user already has.
https://mcp.airtable.com/mcp
Claude Sonnet 4.5
Chat with 60+ AI models on the same workflow — switch to a different model mid-conversation and re-run the same prompt, or use Compare mode to put several side-by-side and balance quality vs. cost.
OAuth (recommended) or an Airtable personal access token. The agent sees only what that user can see.
How models use it and what it is built for.
The server exposes 14 tools. Discovery runs from list_workspaces to list_bases_for_workspace to list_tables_for_base, which is how an agent finds the right table before it reads anything. list_records_for_base queries records, create_records adds up to 10 per request and update_records edits existing ones. Interface pages are readable through list_pages_for_base, list_records_for_page and get_record_for_page. create_base builds a new base with tables and fields. Automations can be listed, drafted, edited and deleted while inactive, but not switched on: a person activates them. That keeps the most consequential step with a human, and it is a sensible boundary for an agent that writes to a shared base.
Typical tools an AI model can call. Exact names vary by version.
Taken from the official Airtable documentation — see Airtable MCP server documentation for the full reference.
Client configuration
Personal access token
Scope the token to the bases the agent needs, and give it read scopes only while you test.
{
"mcpServers": {
"airtable": {
"type": "http",
"url": "https://mcp.airtable.com/mcp",
"headers": {
"Authorization": "Bearer <AIRTABLE_PAT>"
}
}
}
}Copy any of these into MCP Agent Studio after connecting.
In the Content Calendar base, list posts scheduled for next week with no assigned writer.
Mark every task in Q3 Launch whose due date has passed as Overdue.
Create a base for tracking customer interviews with fields for company, date, notes and sentiment.
Draft an automation that emails the owner when a record’s status changes to Blocked.
This is not a single-model product: you get the same MCP connection with 60+ models (Claude, GPT, Gemini, DeepSeek, open-weight, and more), you can switch mid-conversation, and you can open Compare mode to run the same prompt against multiple models at once. The card above is a suggested starting point for this server — not the only choice.
Default pick for Airtable
Claude Sonnet 4.5
Field names and types vary per base, and writes are batched in tens. Sonnet 4.5 inspects the schema first and splits large updates correctly.
Listing tools proves the server is reachable, not that a model can work with it. Evals go further: they read every tool on the server, write a test suite from its real schemas, and run it — code decides pass/fail on the responses (schema conformance, error codes, pagination, result caps) while a scoring model grades plain-English tasks driven through the tools.
Get a pass/fail report per tool with the evidence behind each verdict — and replay the same suite after every schema change. Destructive tools are excluded from the run.
Run evalsOpen MCP Agent Studio with the connection pre-filled. Add your token, pick any of 60+ models, and start chatting — no install required.
Open Agent StudioCommon questions about connecting, scoping and using it safely.
All plans. Standard Airtable API rate limits apply.
No. It can list, draft, edit and delete inactive automations, but a person has to activate them. Script actions are not supported through MCP.
Up to 10 per create request. Larger imports take several calls and count against Airtable’s API rate limits.
It can read interface pages and their records, but it cannot add interface filters or edit an existing interface.
The MCP server blocks development app bases, and it respects existing permissions, so a base the user cannot access returns an error too.