Airtable MCP Server

The Airtable MCP Server is Airtable's official Model Context Protocol server for AI-powered database access. It connects MCP-capable AI clients — including Claude, Claude Code, ChatGPT, Cursor, and Google Gemini — directly to an organisation's Airtable bases. The direct product link to the official developer documentation is https://airtable.com/developers/agents/mcp/getting-started. The open-source CLI wrapper repository is at https://github.com/Airtable/airtable-mcp-cli. According to the provider, the server is available to all users on all plans and requires no own infrastructure.

What is the Airtable MCP Server?

According to the provider, the Airtable MCP Server is a hosted remote server that implements the Model Context Protocol (MCP) and serves as a direct connection between AI agents and structured data stored in Airtable bases. When an agent receives a natural language request — such as "Show me all customers who haven't been contacted in 30 days" or "Add these five new tasks to my project tracker base" — the MCP client calls the Airtable server, which executes the corresponding read or write operation on the configured bases.

The server runs exclusively as a hosted service and is reachable at https://mcp.airtable.com/mcp. No local build step, npm package, or own server infrastructure is required. The repository github.com/Airtable/airtable-mcp-cli contains the official Airtable MCP CLI, a command-line client that fetches commands dynamically from the hosted MCP server at runtime — not the server itself.

Available features

According to the provider, the core toolset of the Airtable MCP Server allows searching bases by name and listing all authorised bases, reading table schemas and field definitions, creating new tables and fields, reading, creating, updating, and submitting records in forms, reading and creating interface pages, and listing, reading, creating, updating, and deleting automations.

Concrete usage examples include retrieving all customers from a CRM base with no contact in the last 30 days, adding multiple new project tasks at once, updating task statuses, creating new bases from a plain-language description, reading interface pages even with interface-only permissions, and managing automations directly in conversation.

Permission model

Per the provider, the Airtable MCP Server's permission model mirrors the authenticated user's existing Airtable permissions. Base owners and creators can modify base schema and edit records. Editors can change and update records. Users with interface-only access can only read pages and data exposed in those interfaces. Commenters and read-only users can read data via MCP but cannot write.

Workspace owners or creators can additionally create new bases via MCP. Base creators without workspace-level permissions cannot create new bases.

Authentication: OAuth and Personal Access Tokens

According to the developer documentation, the Airtable MCP Server supports two authentication methods. OAuth is the recommended path for end-user connections through known clients such as Claude or ChatGPT. After installing the MCP server, a one-time OAuth flow is completed, during which it is possible to choose exactly which bases, apps, and workspaces the AI client may see. Connection permissions can be managed at any time at https://airtable.com/?integrations=thirdParty.

Personal Access Tokens (PATs) are suited for CLI use, scripts, and server-side setups. PATs provide precise, persistent access control without an interactive OAuth flow. Required scopes are data.records:read, data.records:write, schema.bases:read, schema.bases:write, data.recordComments:read, data.recordComments:write, and workspacesAndBases:read. PATs can be managed at https://airtable.com/create/tokens.

Setup: supported clients

Claude and Claude Code offer, according to the provider, a pre-built connector for Airtable that allows a connection in a single step. ChatGPT connects via the Airtable app in the ChatGPT settings. Google Gemini and Gemini Enterprise are also directly supported. For other MCP clients — Cursor, VS Code, and others — manual configuration with the server URL https://mcp.airtable.com/mcp is described in the developer documentation.

Limitations

Per the provider, creating records is limited to 10 records per request. MCP server calls are subject to Airtable's standard API rate limits. Tool names, behaviours, and capabilities may change without prior notice as the server continues to be actively developed.

Security and data access

According to the provider, the Airtable MCP Server processes tool calls and executes the corresponding read or write operations on the authorised Airtable bases. The authenticated user's permissions determine which bases, tables, and records are visible and editable. Since the AI client receives read and write access to Airtable data, it is worth thinking carefully about which bases are enabled for the MCP integration.

Prompt injection via external data sources is a general risk for MCP servers with write access. It is recommended to limit access to only the bases actually needed, test write-capable agents in a safe environment before production use, and regularly review the granted base access permissions. Organisation-wide settings can be controlled by admins; if access to third-party integrations is restricted, an Airtable admin must allowlist the integration.

FAQ

Which plans have access to the Airtable MCP Server? According to the provider, the server is available to all users on all plans. Costs depend on the chosen Airtable subscription and AI client — not on the MCP server itself.

Do I need to install anything locally? No. The Airtable MCP Server runs exclusively as a hosted remote service. No local process, npm package, or own infrastructure is required. The optional CLI (@airtable/mcp-cli) enables terminal use but does not replace the server itself.

How does the Airtable MCP CLI differ from the MCP Server? The MCP Server (mcp.airtable.com/mcp) is the actual hosted server that AI clients communicate with. The Airtable MCP CLI (@airtable/mcp-cli) is a command-line tool that fetches commands dynamically from the server at runtime, enabling direct use from the terminal without a separate MCP client.

An Airtable account (all plans), data in Airtable bases, and an MCP-capable AI client such as Claude, Claude Code, ChatGPT, or Cursor. For organisational accounts, admin approval for third-party integrations may be required.

For Claude and Claude Code: use the pre-built connector — connect in a single step via the Airtable integrations page. For other clients: add server URL https://mcp.airtable.com/mcp to the MCP configuration and use the OAuth flow or a PAT for authentication.

claude mcp add --transport http airtable https://mcp.airtable.com/mcp

OAuth flow (recommended for end-user connections, Dynamic Client Registration supported) or Personal Access Token (PAT) for CLI, scripts, and non-interactive environments.

Access permissions mirror existing Airtable permissions. Owner/Creator: modify schema and records. Editor: change records. Interface-only: read interface pages only. Commenter/Read-only: read only.

The server receives tool calls and executes read or write operations on the authorised Airtable bases. Data flows directly between the AI client and Airtable infrastructure.

AI clients receive read and write access to the enabled bases. Enable only the bases actually needed. Prompt injection via external data sources is a general MCP risk with write access; test write-capable agents before production use.