airtable-mcp

Self-discovery

Tools are fetched from the MCP server at runtime, so the CLI never has a hardcoded command list. Discover what's available:

airtable-mcp tools            # human-readable list
airtable-mcp tools --json     # machine-parseable list
airtable-mcp <tool> --help    # show flags and descriptions for a tool

Run airtable-mcp tools before assuming a tool exists. Tool names, arguments, and output shapes can change between server releases without a CLI update.

Install

npm install -g @airtable/mcp-cli

Auth

The CLI needs an Airtable personal access token (PAT). Two paths:

Environment variable (preferred for scripts/agents):

export AIRTABLE_TOKEN=pat_xxx

Interactive configure (stores token in ~/.airtable/cli.json with 0600 permissions):

airtable-mcp configure

Create tokens at https://airtable.com/create/tokens. Ensure the token has the scopes required by the tools being called.

AIRTABLE_TOKEN takes precedence over saved profiles when no --profile flag is set. Never log or echo tokens.

Quick reference

Tool names use hyphens on the CLI (list-records) but underscores in MCP (list_records). The CLI translates automatically.

Workflow

1. Auth — set AIRTABLE_TOKEN or run airtable-mcp configure
2. Discover — run airtable-mcp tools to see available tools
3. Inspect — run airtable-mcp <tool> --help for flags and descriptions
4. Check access — in tools --json output, check the access field: read-only, write, or destructive. Confirm with the user before running destructive tools.
5. Execute — run airtable-mcp <tool> --flagName value

Output & automation

• Default output is formatted JSON to stdout. Status messages go to stderr.
• --json on tools gives a JSON array of {name, title, access}.
• -q / --quiet suppresses stderr status messages (cache warnings, etc).
• --output raw returns the raw server response text instead of parsed JSON.
• --input - reads tool arguments as a JSON object from stdin, bypassing flag parsing.
• Exit codes: 0 success, 1 error (auth, tool failure, not found), 2 usage error (bad flags, bad input).

Common tasks

Find a base and list its tables:

airtable-mcp search-bases --searchQuery "Project Tracker" -q
airtable-mcp list-tables-for-base --baseId appEXAMPLEbase001 -q

List records with specific fields:

airtable-mcp list-records-for-table \
  --baseId appEXAMPLEbase001 --tableId tblEXAMPLEtable01 \
  --fieldIds '["Name","Status"]' --pageSize 10 -q

Filter records — filters use structured JSON, not formula strings. Wrap conditions in an operands array; the top-level operator defaults to and if omitted:

airtable-mcp list-records-for-table \
  --baseId appEXAMPLEbase001 --tableId tblEXAMPLEtable01 \
  --filters '{"operator":"and","operands":[{"operator":"=","operands":["Status","Done"]}]}' -q

For select fields, filter by choice ID (from get-table-schema), not the display name. The airtable-filters skill covers compound filters, date filters, and operator-by-field-type details.

Search records — use search-records for free-text/fuzzy queries on large tables. Use list-records-for-table with --filters when filtering by exact field values:

airtable-mcp search-records \
  --baseId appEXAMPLEbase001 --table tblEXAMPLEtable01 \
  --query "acme" --fields '["Name","Notes"]' -q

Pass --fields ALL_SEARCHABLE_FIELDS to search across every indexed field. Date, rating, checkbox, and button fields are not searchable.

Update records — complex args are easier via --input -:

echo '{"baseId":"appEXAMPLEbase001","tableId":"tblEXAMPLEtable01","records":[{"id":"recEXAMPLErecord1","fields":{"fldEXAMPLEfield01":"Done"}}]}' \
  | airtable-mcp update-records-for-table --input - -q

Select field values are returned as objects ({"id":"sel...","name":"Done"}) but must be written as plain strings ("Done"). Record field keys in create/update currently require field IDs (fldEXAMPLEfield02) — use get-table-schema to resolve names to IDs before writing. Note that fieldIds, sort, and filters accept both names and IDs.

Gotchas