Airtable MCP server
Maintained Airtable MCP server: let an agent inspect base schemas, then read, search, and write records, tables, fields, and comments.
airtable-mcp-server is a popular, actively maintained MCP server that gives an AI agent read and write access to Airtable. Airtable is a spreadsheet-database hybrid where teams keep CRMs, content calendars, product catalogs, and project trackers. The hard part for an agent is that it cannot guess your bases, tables, and field types — so this server leads with schema inspection: it lists the bases you can reach, lists and describes tables (at configurable detail levels), and surfaces field definitions, so the agent learns the structure before touching data. From there it can query records with formulas, full-text search across fields, fetch single records, and create, update, or delete records.
The server runs locally over stdio via npx (npx -y airtable-mcp-server) and authenticates with an Airtable personal access token supplied as AIRTABLE_API_KEY. You scope the token's permissions at creation time — for example schema.bases:read and data.records:read for a read-only setup, adding the :write scopes only when you want the agent to modify data, schema, or comments. Beyond record CRUD it can create and update tables and fields and manage record comments, making it a full-surface tool for automating Airtable maintenance from your editor or chat.
Quick install
Copy-paste configs are provided for all 8 supported clients. Pick your client below.
Add to ~/.claude.json
{
"mcpServers": {
"airtable": {
"command": "npx",
"args": [
"-y",
"airtable-mcp-server"
],
"env": {
"AIRTABLE_API_KEY": "<AIRTABLE_API_KEY>"
}
}
}
}claude mcp add airtable -- npx -y airtable-mcp-serverAvailable tools
| Tool | Description |
|---|---|
| list_records | List records from a table, with optional maxRecords and a filterByFormula expression. |
| search_records | Search for records containing specific text across all or selected fields. |
| list_bases | List all accessible Airtable bases with their ID, name, and permission level. |
| list_tables | List all tables in a base at a configurable detail level (identifiers only or full schema). |
| describe_table | Get detailed information about a single table, including its fields and views. |
| get_record | Get a specific record by its ID from a table. |
| create_record | Create a new record in a table with the given field values. |
| update_records | Update one or multiple existing records in a table. |
| delete_records | Delete one or more records from a table. |
| create_table | Create a new table in a base with defined fields. |
| update_table | Update a table's name or description. |
| create_field | Add a new field to an existing table. |
| update_field | Update a field's name or description. |
| create_comment | Post a comment on a record, with optional threading. |
| list_comments | List the comments on a record, with pagination. |
Required configuration
- AIRTABLE_API_KEYRequired
Airtable personal access token (format pat...), scoped to the bases and permissions you grant. Required.
What you can do with it
Query and update a base without leaving your editor
The agent lists your bases, describes the relevant table to learn its fields, then runs list_records with a filterByFormula or search_records to find what it needs — and updates rows with update_records once you confirm.
Automate base maintenance and schema changes
Hand the agent a setup task and it can create_table, add columns with create_field, and seed rows with create_record, then leave context for collaborators using create_comment — turning a manual Airtable chore into one prompt.
FAQ
- Is it free?
- Yes. airtable-mcp-server is open source (MIT) and free to run via npx. It uses your Airtable account, so Airtable's own plan limits and pricing apply to the bases and API usage.
- Does it support remote/OAuth?
- The canonical mode is local stdio via npx with an Airtable personal access token in AIRTABLE_API_KEY; there is no OAuth flow. Access is controlled by the token's scopes — grant read-only scopes for a safe read setup, or add write scopes to let the agent modify data and schema. It can also run as an HTTP server for remote use.