Airtable MCP Server

Model Context Protocol server that gives LLMs full read/write access to your Airtable bases.

Free· Open source (MIT) and free to run. You still need an Airtable account and personal access token; Airtable itself is Free / Team $20 per user/mo / Business $45 per user/mo / Enterprise (custom).MCP Servers

Visit website →

Best for

Developers and ops teams who already run their CRM, content calendar or project tracker in Airtable and want an LLM assistant that can actually read and edit those bases through Claude Desktop, Cursor or Claude Code.

Skip if

Enterprise deployments that need SSO, audit logs and vendor SLAs around the Airtable integration, or non-technical users who cannot install an npm package and edit an MCP config file.

Airtable MCP Server is an open-source Model Context Protocol implementation by Adam Jones that plugs Airtable into any MCP-compatible LLM client — Claude Desktop, Claude Code, Cursor, Cline, VS Code and the rest of the growing MCP ecosystem. Once installed, the assistant can introspect your workspace (list_bases, list_tables, describe_table), read data (list_records, search_records, get_record with filterByFormula support), and mutate it (create_record, update_records, delete_records) — plus, if you grant the right token scopes, edit schema itself with create_table, update_table, create_field and update_field, and read or post record comments. The server is a tiny TypeScript package published to npm; the recommended install is a one-liner npx invocation wired up by the author's install-mcp helper, which generates the correct JSON config for whichever client you use. You authenticate with an Airtable personal access token and choose scopes granularly, so read-only agents stay read-only. A stateless HTTP transport (MCP_TRANSPORT=http) is available for remote clients, with an explicit and prominent warning that it has no built-in auth and must sit behind a reverse proxy — the maintainer is up-front about the DNS-rebinding threat model rather than pretending localhost binding is a security boundary. Typical workflows: letting Claude answer questions grounded in a CRM base, having an agent triage or enrich inbound leads, generating scheduled reports off project trackers, or bootstrapping brand-new tables and fields from a natural-language spec. It is the de-facto community Airtable MCP: 400+ GitHub stars, active maintenance, and covers essentially the full Airtable REST surface most teams need.

Editor's take

This is the Airtable MCP server we reach for. The tool surface is complete, the install-mcp helper removes the usual JSON-config pain, and the README's frank DNS-rebinding warning tells you the maintainer thinks about the boring safety stuff. Just keep the token scoped to what the agent actually needs — write access is a footgun.

— The AI Tool Bible editorial team

Pros

✅ Broad tool coverage — records, schema inspection, table/field creation, and record comments in one server.
✅ Works with every major MCP client (Claude Desktop, Claude Code, Cursor, Cline, VS Code) via a generated install config.
✅ Fine-grained scope control through Airtable personal access tokens lets you ship a strictly read-only agent.
✅ Open source under MIT, zero runtime cost, distributed as a small npm package you can pin or fork.
✅ Supports filterByFormula and a dedicated search_records tool so agents can query without downloading whole tables.
✅ Honest, prominent security warning about the HTTP transport rather than a silent footgun.

Cons

⚠️ HTTP transport ships with no authentication — safe defaults require a reverse proxy you have to configure yourself.
⚠️ No batching primitives beyond Airtable's own limits, so large writes can be slow and quota-hungry.
⚠️ Airtable API rate limits (5 req/sec/base) still apply and can bottleneck chatty agents.
⚠️ Requires a personal access token with real write power; a compromised MCP client can delete records or alter schema.
⚠️ Community project maintained largely by one developer — response times to issues will vary.

Use cases

Natural-language CRM queries against AirtableLLM-driven lead triage and enrichmentAutomated content calendar updatesBootstrapping new bases, tables and fields from a specRead-only reporting agentsProject tracker status summariesBulk record cleanup via chatThreaded comments and mentions from an assistant