SuperAPI AI Directory

airtable-mcp-server

Airtable MCP Server is a Model Context Protocol server that provides AI systems with read and write access to Airtable databases. It enables LLMs to inspect database schemas and to read, create, update, and delete records, as well as manage tables, fields, and comments. Through a standardized MCP interface, AI clients (such as Claude Code, Claude Desktop, Cursor, Cline, VS Code, etc.) can seamlessly connect to Airtable and perform data operations. The server requires configuration of an Airtable personal access token and supports both stdio and HTTP transport modes, making it suitable for local or remote integration scenarios.

Open website

About this tool

Airtable MCP Server is a server that implements the Model Context Protocol (MCP), designed to enable AI systems (such as large language models) to interact with Airtable databases. It provides read and write access to Airtable databases, enabling LLMs to inspect database schemas and to read and write records.

Main Capabilities

The server provides a set of tools (Tools) for operating on Airtable resources:

Record Operations

  • list_records: List records in a specified Airtable table. Supports limiting the number of returned records via maxRecords (default 100), and can filter by formula using filterByFormula .
  • search_records: Search for records containing specific text. You can specify the field ID to search, and support maxRecords to limit the number of returned records.
  • get_record: Get a specific record by record ID.
  • create_record: Create a new record in a table; fields and values must be provided.
  • update_records: Update one or more records; record IDs and the fields to update must be provided.
  • delete_records: Delete one or more records; an array of record IDs must be provided.

Base and Table Management

  • list_bases: List all accessible Airtable bases, returning base ID, name, and permission level.
  • list_tables: List all tables in a specific base. The level of detail can be controlled via detailLevel (tableIdentifiersOnly、identifiersOnly or full), returning table ID, name, description, fields, and views.
  • describe_table: Get detailed information about a specific table; the parameters and return format are similar to list_tables , but only for a single table.
  • create_table: Create a new table in a base; the table name and an array of field definitions (including name, type, description, options) must be provided.
  • update_table: Update a table's name or description.

Field Management

  • create_field: Create a new field in a table; name and type must be provided, with optional description and field-specific options.
  • update_field: Update a field's name or description.

Comment Management

  • create_comment: Create a comment on a record; text must be provided, with an optional parent comment ID for threaded replies.
  • list_comments: List comments on a record, supporting pagination (pageSize and offset), returning an array of comments along with author, text, timestamp, reactions, and mention information, sorted from newest to oldest.

Usage

  1. Installation and Configuration: Follow the instructions for install-mcp, which will generate the correct configuration for your MCP client (such as Claude Code, Claude Desktop, Cursor, Cline, VS Code, etc.).
  2. Obtain an Access Token: You need to create an Airtable personal access token. When creating it, the following scopes must be included:schema.bases:read and data.records:read(optional:schema.bases:write、data.records:write、data.recordComments:read、data.recordComments:write), and ensure the token has permission to access the target base. The token format is similar to pat123.abc123(actually longer). Set it as the environment variable AIRTABLE_API_KEY(replacing the placeholder in the generated configuration).
  3. Run modes:
    • Default stdio transport: suitable for MCP clients on the same machine, does not open a port.
    • HTTP mode: for remote MCP clients. After startup, it provides a stateless HTTP server at http://localhost:3000/mcp .

Use cases

  • Let AI assistants directly query, analyze, and manipulate data in Airtable without manual export and import.
  • Automate data entry, update, and cleanup processes.
  • Provide LLMs with structured data access to enhance the accuracy and contextual relevance of their answers.
  • Integrate Airtable as a data backend in a development environment for rapid prototyping through AI.

Limitations and security considerations

  • The HTTP transport mode has no built-in authentication, and binding to localhost or a private network does not prevent browser cross-origin attacks: a malicious website can use DNS rebinding to make a visitor's browser send requests to http://localhost:3000/mcp and invoke tools—including using the server's Airtable token to read, write, and delete records.
  • HTTP mode should only be run in environments where untrusted callers (including browsers on the same machine or network) cannot access /mcp without authentication. In practice, place it behind a reverse proxy or gateway that requires credentials (such as an Authorization header, which browsers will not attach cross-origin).
  • If you only want to use it with an MCP client on the same machine, use the default stdio transport, which does not open a port.

Development and contributions

Pull requests are welcome on GitHub. Development steps: install Git and Node.js; clone the repository; use npm install to install dependencies; run npm run test to execute tests; use npm run build to build; you can use npm run build:watch to automatically build after editing src/index.ts , making it convenient to reload the Claude Desktop (Ctrl/Cmd+R) app to apply changes after saving. Versions follow semantic versioning; use npm version <major | minor | patch> to bump the version, then git push --follow-tags.

Note: The above information is based on the given materials; details not provided (such as specific pricing, platform limitations, etc.) are not yet available.