airtable-user-mcp
Airtable Formula
Formula editor, MCP server, and AI skills for VS Code
Not affiliated with Airtable Inc. This is a community-maintained project.
Experimental — This project is under active development and not intended for production use. APIs, tools, and behavior may change without notice.
Why this exists
Airtable's public Web API has never exposed some of the most common tasks builders actually need: creating a formula field, tweaking a view's filter set, installing an extension, or validating a formula before it breaks production. The official Airtable MCP server is a thin wrapper over that same REST API, so it inherits every one of those gaps.
airtable-user-mcp is an add-on to the official Airtable MCP, not a replacement. It uses Airtable's own internal API (the one the web UI uses) to cover exactly the surface area the REST API can't reach. Register both servers in your AI client and your assistant gets the full Airtable automation experience — records over HTTP via the official MCP, plus schema, formulas, views, and extensions via this one.
Related MCP server: Airtable MCP Server
What airtable-user-mcp adds on top of the official Airtable MCP
This is a coverage map, not a "pick one" decision — the two servers are complementary and designed to run side-by-side.
Capability | Official Airtable MCP | airtable-user-mcp |
Total tools | ~17 | 62 |
Auth model | Personal Access Token or OAuth, per-scope setup | Log in once with your normal Airtable account (SSO/2FA supported) |
Transport | HTTP (remote) | stdio (local, private) |
Data never leaves your machine | ❌ Requests go through | ✅ Runs locally against Airtable's API |
Schema read (bases, tables, fields, views) | Partial (no view config) | Full — filters, sorts, groups, visibility, row height, descriptions |
Create formula fields | ❌ | ✅ |
Create rollup fields | ❌ | ✅ |
Create lookup / multipleLookupValues fields | ❌ | ✅ |
Create count fields | ❌ | ✅ |
Update formula text of an existing field | ❌ | ✅ |
Validate a formula before applying | ❌ | ✅ |
Rename / duplicate / safely delete fields | Partial (no duplicate, no dependency summary) | ✅ with |
Create views (grid/form/kanban/calendar/gallery/gantt/list) | ❌ (API has no create-view endpoint) | ✅ |
Set/append view filters (nested AND/OR) | ❌ | ✅ |
Set view sorts | ❌ | ✅ |
Set view grouping | ❌ | ✅ |
Change column order | ❌ | ✅ |
Show/hide columns in a view | ❌ | ✅ |
Change row height | ❌ | ✅ |
Duplicate a view with its full configuration | ❌ | ✅ |
View descriptions, cell wrap, covers, color config, calendar dates, frozen columns | ❌ | ✅ |
Sidebar sections (create, rename, move, delete) | ❌ | ✅ |
Record templates (create, pre-fill, duplicate, apply, delete) | ❌ | ✅ |
Form metadata (description, redirect, attribution, branding) | ❌ | ✅ |
Extension / block management (install, enable, rename, duplicate, remove) | ❌ | ✅ |
Create dashboard pages | ❌ | ✅ |
Tool profiles & per-tool toggles | ❌ | ✅ read-only / safe-write / full / custom |
| ❌ Explicitly disallowed | ✅ |
Destructive-action safety guards | Relies on token scopes | ✅ |
Batch record create limit | 10 / request | Uses the same Airtable limit; no added restriction |
VS Code / Cursor / Windsurf / Cline / Amp one-click install | Manual JSON edit per IDE | ✅ One click via the companion extension |
Formula editor with IntelliSense | ❌ | ✅ (VS Code extension) |
Credentials storage | You manage the PAT | OS keychain, auto-refresh |
Plan requirement | Airtable plan with API access + token scopes | Any plan you can log into |
Price | Free | Free, MIT |
Sources: Airtable's official MCP docs, Airtable Web API reference, and the UNSUPPORTED_FIELD_TYPE_FOR_CREATE rollup thread.
Use both MCPs together
airtable-user-mcp is additive. Register the official Airtable MCP following Airtable's setup guide, then add this one alongside it in the same mcpServers block:
{
"mcpServers": {
"airtable-user-mcp": {
"command": "npx",
"args": ["-y", "airtable-user-mcp"]
}
}
}Your MCP client will expose every tool from both servers. The two entries are independent — rename the keys (airtable, airtable-official, airtable-user-mcp, etc.) however makes sense for your workflow.
What's In This Repo
This monorepo ships two products from one source tree:
Product | Install | |
Airtable Formula — VS Code extension | ||
airtable-user-mcp — Standalone MCP server |
|
Demo
Features
VS Code Extension
Formula Editor — Syntax highlighting, IntelliSense, beautify / minify for
.formulafilesMCP Server — One-click MCP registration for multiple IDEs
AI Skills — Auto-install Airtable-specific skills, rules, and workflows for AI coding assistants
Airtable Login — Credentials in OS keychain, browser-based auth with auto-refresh
Dashboard — React webview with Overview, Setup, and Settings tabs
MCP Server (62 Tools)
Manage Airtable bases with capabilities not available through the official REST API:
Category | Tools | Highlights |
Schema Read | 9 | Full schema inspection — bases, tables, fields, views, sidebar sections, record templates |
Table Management | 3 | create / rename / delete tables |
Field Management | 8 | Create formula / rollup / lookup / count fields, validate formulas, update descriptions, delete |
View Configuration | 20 | Filters, sorts, grouping, columns, freezing, row height, covers, color rules, calendar dates, create / duplicate / rename / delete |
Sidebar Sections | 4 | Create, rename, move-into-section, delete (auto-promotes contained views to ungrouped) |
Record Templates | 8 | Create / rename / describe / set cells / set columns / duplicate / apply / delete saved row scaffolds |
Form Metadata | 2 | Description, redirect URL, attribution, copy-to-respondent, branding (legacy form views) |
Extension Management | 7 | Create, install, enable/disable, rename, duplicate, remove extensions |
Tool Management | 1 | List profiles, switch profile, toggle tools/categories (meta-tool, always enabled) |
See the full tool reference in packages/mcp-server/README.md.
Supported IDEs
The extension auto-configures MCP for all major AI-enabled editors:
Claude Desktop | Claude Code | Cursor | Windsurf | Cline | Amp |
Don't use VS Code? Use the standalone MCP server directly:
npx airtable-user-mcpFind Us
Registry | Link |
VS Code Marketplace | |
npm | |
Open VSX | |
MCP Registry | |
Glama | |
PulseMCP | |
MCP.so | |
Video demo — managing views, computed fields & extensions with Claude Code |
Requirements
VS Code ^1.100.0 (or any fork exposing the
McpServerDefinitionProviderAPI)Node.js — bundled via the VS Code runtime; no separate install needed
Google Chrome (or Edge / Chromium) — the Airtable login flow uses Patchright in headless mode. Falls back to
msedgeon Windows andchromiumon Linux. The extension shows an actionable warning if no supported browser is detected.
Development
This is a pnpm monorepo.
Package | Description |
| VS Code extension host (TypeScript + tsup) |
| React dashboard webview (Vite + Tailwind v4) |
| Shared types and message protocol |
|
|
| Build tooling (esbuild bundler, dep vendoring) |
pnpm install # install all packages
pnpm build # build shared → webview → mcp bundle → extension
pnpm package # build + create airtable-formula-X.Y.Z.vsix
pnpm test # run all unit tests
pnpm dev # start webview dev server (browser preview)How the MCP server is bundled: scripts/bundle-mcp.mjs esbuilds packages/mcp-server/src/ into packages/extension/dist/mcp/. Then scripts/prepare-package-deps.mjs vendors patchright, patchright-core, and otpauth into dist/node_modules/ before vsce package runs. The VSIX is fully self-contained.
Support This Project
This project is built and maintained with the help of AI coding tools. If you find it useful and want to support continued development (new tools, updates, bug fixes), you can contribute by gifting Claude Code credits — the primary tool used to build this project.
Interested? Open an issue or reach out to discuss feature requests and sponsorship.
License
Available Tools
73 toolsapply_record_templateA
Apply (instantiate) a record template to create a new record pre-filled with the template's field values. Returns the new record data.
Inputs: appId (required), templateId (required), debug.
apply_view_sortsAIdempotent
Apply sort conditions to a view. Default mode replaces all existing sorts — pass an empty array with operation="replace" to clear. Use operation="append" to add new sorts on top of the view's existing sort stack without rewriting them.
Inputs: appId (required), sorts (required), viewId (required), debug, +1 more.
create_extensionA
Create a new extension (block) in an Airtable base. Returns the block ID needed for installation. Use this to register custom extensions before installing them.
Inputs: name (required), appId (required), releaseId (required), debug.
create_extension_dashboardB
Create a new extension dashboard page in a base. Extensions are installed onto dashboard pages.
Inputs: name (required), appId (required), debug.
create_fieldA
Create a new field in an Airtable table. Supports all field types including computed fields (formula, rollup, lookup, count) that are not available via the official API.
FIELD TYPES (fieldType parameter): Supported names: "text", "multilineText", "number", "checkbox", "date", "singleSelect", "multipleSelects", "rating", "formula", "rollup", "lookup", "count" Friendly aliases (auto-normalized): "url" → type: "text" with validatorName = "url" "email" → type: "text" with validatorName = "email" "phone" / "phoneNumber" → type: "text" with validatorName = "phoneNumber" "dateTime" → type: "date" with isDateTime: true
TYPE OPTIONS by fieldType: formula: { formulaText: "..." } rollup: { relationColumnId: "fldLINK", foreignTableRollupColumnId: "fldTARGET", formulaText: "SUM(values)" } (formulaText is REQUIRED — e.g. "SUM(values)", "COUNTA(values)", "IF(OR(values='X'),1,0)")) (old keys fieldIdInLinkedTable/recordLinkFieldId are auto-translated for backward compat) lookup: { relationColumnId: "fldLINK", foreignTableRollupColumnId: "fldTARGET" } (old keys fieldIdInLinkedTable/recordLinkFieldId are auto-translated for backward compat) count: { recordLinkFieldId } number (integer): { format: "integer", negative: false } number (currency): { format: "currency", symbol: "$", precision: 2, negative: false } number (percent): { format: "percentV2", precision: 2, negative: false } date / dateTime: { dateFormat: "Local"|"us"|"european"|"iso"|"friendly", timeFormat: "12hour"|"24hour", timeZone: "UTC"|"client"|, shouldDisplayTimeZone: true|false, isDateTime: true (auto for dateTime) } singleSelect: { choices: [{ name: "Option A", color: "blue" }], default: "selXXX" } multipleSelects: { choices: [{ name: "PC", color: "blue" }, { name: "Xbox", color: "cyan" }], default: ["selXXX"] } text / multilineText / checkbox / rating: omit typeOptions entirely — passing {} causes a 422
SELECT CHOICES:
Pass choices as an array [{ name, color? }] or as an object { selXXX: { name, color? } }.
The client auto-adds id inside each choice value, generates choiceOrder, and sets disableColors: false.
Color names (confirmed): "blue", "cyan", "teal", "green", "yellow", "orange", "red", "pink", "purple", "gray".
"default" sets the pre-selected value: string ID for singleSelect, array of IDs for multipleSelects.
To add/remove choices without losing existing ones, call get_table_schema first and include ALL choices in the update.
Inputs: name (required), appId (required), tableId (required), fieldType (required), +4 more.
create_formula_fieldA
Create a new formula field — shorthand for create_field with type "formula". Use create_field for all other field types (singleSelect, rollup, number, etc.). Returns { columnId }.
Inputs: name (required), appId (required), tableId (required), debug, +2 more.
create_recordsA
Create one or more records in a table. Each item supplies cellValuesByColumnId (computed fields are read-only and must be omitted). Returns created record IDs. Per-row isolation: a failing row is reported, not fatal.
Inputs: appId (required), records (required), tableId (required), debug, +1 more.
create_record_templateA
Create a new record template for a table. Returns the generated templateId (rtp-prefixed). After creating, use set_record_template_cell to pre-fill field values.
Inputs: appId (required), tableId (required), name, debug.
create_tableA
Create a new table in an Airtable base. Returns the generated table ID. The table starts with default fields (Name, Notes, Attachments, Status, etc.) — use list_fields after creation to inspect them.
Inputs: name (required), appId (required), debug.
create_viewB
Create a new view in an Airtable table. Optionally copy configuration from an existing view. View types: "grid", "form", "kanban", "calendar", "gallery", "gantt", "levels" (list view).
Inputs: name (required), appId (required), tableId (required), type, +2 more.
create_view_sectionA
Create a new sidebar section in a table. Returns the new section ID (vsc-prefixed). Use move_view_to_section to populate it with views.
Inputs: name (required), appId (required), tableId (required), debug.
delete_fieldADestructive
Delete a field from an Airtable table. Requires fieldId AND expectedName as a safety guard — deletion is refused if the name does not match. ⚠️ Irreversible: deleted field data is permanently lost and cannot be recovered. Always checks downstream dependencies first (formula fields, lookups, rollups referencing this field); returns dependency info without deleting unless force=true.
Inputs: appId (required), fieldId (required), expectedName (required), debug, +1 more.
delete_fieldsADestructive
Delete multiple fields from an Airtable table in a single call. Each entry requires fieldId and expectedName as a safety guard (deletion is refused if names do not match). Fields are processed sequentially and all are attempted even if some fail — partial results are always returned. Optionally writes a JSON checkpoint file after each deletion so the batch can be resumed if interrupted.
Inputs: appId (required), fields (required), debug, force, +1 more.
delete_recordsADestructive
Delete one or more records from a table in a single batch call. The returned deleted count equals rowIds.length (optimistic) — already-deleted rows are silently skipped by the server.
Inputs: appId (required), rowIds (required), tableId (required), debug, +1 more.
delete_record_templateADestructive
⚠️ DESTRUCTIVE — Permanently delete a record template. This cannot be undone.
Inputs: appId (required), templateId (required), debug.
delete_tableADestructive
Delete a table from an Airtable base. Requires both tableId AND the expected table name as a safety guard — refuses to delete if the name does not match. Airtable rejects deleting the last remaining table in a base.
Inputs: appId (required), tableId (required), expectedName (required), debug.
delete_viewADestructive
Delete a view from a table. Cannot delete the last remaining view in a table.
Inputs: appId (required), viewId (required), debug.
delete_view_sectionADestructiveIdempotent
Delete a sidebar section. Views inside the section are NOT deleted — Airtable auto-promotes them to ungrouped at the table-level position the section used to occupy. Verified 2026-04-30.
Inputs: appId (required), sectionId (required), debug.
download_base_formulasADestructiveIdempotent
Download ALL formula fields from a base to local .formula files, organized into per-table subfolders. Field refs are resolved to real field names ({Field Name}, Airtable's native syntax) so files are readable and upload back unchanged. Each file includes a # AT: header with appId, tableId, fieldId, fieldName, description, and resultType. Tables with no formula fields are silently skipped. outputDir defaults to the current working directory when omitted.
Inputs: appId (required), debug, outputDir.
download_formula_fieldADestructiveIdempotent
Download the formula text of a formula field to a local file. Field refs are resolved to real field names ({Field Name}, Airtable's native syntax) so the file is readable and can be uploaded back unchanged. Writes a .formula file with a # AT: metadata header (appId, tableId, fieldId, fieldName) so the file can later be uploaded back with update_formula_field or the VS Code right-click command. When outputPath is omitted, returns the formula text without writing a file.
Inputs: appId (required), fieldId (required), debug, outputPath.
duplicate_extensionB
Duplicate an installed extension on a dashboard page.
Inputs: appId (required), pageId (required), sourceInstallationId (required), debug.
duplicate_fieldB
Duplicate (clone) a field in a table. Optionally also duplicate the cell values.
Inputs: appId (required), tableId (required), sourceFieldId (required), debug, +1 more.
duplicate_recordsA
Duplicate one or more existing records within a table. Creates exact copies of the specified source records in the same table and view. Returns the new record IDs.
Inputs: appId (required), viewId (required), tableId (required), sourceRowIds (required), +1 more.
duplicate_record_templateA
Duplicate a record template within the same or a different table. Returns the new template ID.
Inputs: appId (required), tableId (required), sourceTemplateId (required), debug, +1 more.
duplicate_viewA
Duplicate an existing view with all its configuration (filters, sorts, field visibility, etc).
Inputs: appId (required), newName (required), tableId (required), sourceViewId (required), +1 more.
get_base_schemaARead-onlyIdempotent
Get the full schema of an Airtable base — all tables, fields (with typeOptions), and views in one call. Use this when you need fields or views; use list_tables when you only need table names/IDs (faster, lighter). Returns { tables: [...] }.
Inputs: appId (required), debug.
get_table_schemaARead-onlyIdempotent
Get the full schema for a single table — all fields (with typeOptions) and views. Use instead of get_base_schema when you only need one table (faster, less context). Use list_fields when you need fields only without view data. Returns { id, name, fields: [...], views: [...] }.
Inputs: appId (required), tableIdOrName (required), debug.
get_viewARead-onlyIdempotent
Read a view's live configuration from the base. Returns filters, sorts, groupLevels, columnOrder (rich per-column visibility + width), frozenColumnCount, colorConfig, metadata (view-type specific, e.g. gallery cover, calendar date field), rowHeight, description. Use this before update_view_filters / apply_view_sorts / update_view_group_levels to audit current state and choose between replace and append modes.
Data source: internally hits /v0.3/table/{tableId}/readData with includeDataForViewIds=[viewId]. The application/read endpoint alone does NOT return filter/sort/group state — that's why the update tools need either "append" mode or a prior get_view call to merge safely.
Fields:
filters: { filterSet: [...], conjunction: "and"|"or" } | null
sorts: [{ id, columnId, ascending }] | null (stored as lastSortsApplied internally)
groupLevels: [{ id, columnId, order, emptyGroupState }] | null
columnOrder: [{ columnId, visibility, width? }]
visibleColumnOrder: [columnId] — derived from columnOrder for convenience
metadata: type-specific config (gallery.coverColumnId, calendar.dateColumnId, etc.)
Inputs: appId (required), viewId (required), debug.
install_extensionA
Install an extension onto a dashboard page. Requires a block ID (from create_extension) and a page ID (from create_extension_dashboard).
Inputs: name (required), appId (required), pageId (required), blockId (required), +1 more.
list_fieldsARead-onlyIdempotent
List fields in a table — returns id, name, and type per field (lightweight by default). Use instead of get_table_schema when you need fields only (no view data). Set includeOptions=true to also return each field's full typeOptions (can be very large on wide tables — prefer the default for name/id/type lookups). Use fieldType or nameContains filters on large tables to reduce context size. Returns [{ id, name, type, typeOptions? }].
Inputs: appId (required), tableIdOrName (required), debug, fieldType, +2 more.
list_record_templatesARead-onlyIdempotent
List all record templates for a table. Templates are embedded in the base scaffolding data. If the templates array is empty, pass debug:true and inspect the raw response to locate the templates key — the API path may vary by base.
Inputs: appId (required), tableId (required), debug.
list_tablesARead-onlyIdempotent
List all tables in a base with their IDs and names — lightweight scaffolding call (no field data). Use this when you only need table IDs/names; use get_base_schema or get_table_schema when you also need fields or views. Returns [{ id, name }].
Inputs: appId (required), debug.
list_viewsARead-onlyIdempotent
List all views in a specific table with their IDs, names, and types.
Inputs: appId (required), tableIdOrName (required), debug.
list_view_sectionsARead-onlyIdempotent
List all sidebar sections for a table. Sections are user-organized groupings of views in the Airtable left sidebar (e.g. "🚀 Posting workflow", "🗑️ Sold workflow"). Returns each section's id, name, and the views inside it. The table-level tableViewOrder is a mixed list of view IDs and section IDs at the top level — when a view is inside a section, it appears in that section's viewOrder, NOT in the table's.
Inputs: appId (required), tableIdOrName (required), debug.
manage_daemonAIdempotent
Inspect and control the MCP daemon this server runs in. action="status" is the diagnostic to reach for FIRST when tools start failing: it reports whether a daemon is running and whether YOU are it, the transport, uptime, version/provenance, the tunnel URL, and — the part nothing else exposes — the live session state (sessionDead, the last circuit-breaker trip INCLUDING Airtable's own response body, and the browser/auth busy queue). That is how you tell "daemon gone" from "session dead" from "browser busy" instead of guessing. Control actions: start (idempotent; attaches to a healthy daemon rather than starting a second one), restart, stop (writes a sentinel so the VS Code extension does not silently respawn it), tunnel_enable, tunnel_disable, token_rotate. stop/restart answer first and exit afterwards, so this call returns normally and the daemon goes away a moment later. token_rotate and tunnel_* are loopback-only and are refused for callers arriving through the tunnel; status is returned to them with host-identifying fields blanked. The bearer token is never returned to anyone. Interactive tunnel setup (cloudflared login, creating a named tunnel) is deliberately NOT here — use the CLI or the VS Code dashboard.
Inputs: action (required), domain, reason, provider.
manage_toolsAIdempotent
Control which tools are available. Actions: list_profiles, switch_profile, get_tool_status, toggle_tool, toggle_category. Use this to switch between read-only, safe-write, full, or custom profiles, or enable/disable individual tools.
Active profile: "full" — all tools enabled.
Inputs: action (required), tool, enabled, profile, +1 more.
move_overall_columnsAIdempotent
Move one or more columns to a new position in the overall index (visible + hidden). Sibling of move_visible_columns. Index 0 is the leftmost column in the underlying full order.
Inputs: appId (required), viewId (required), columnIds (required), targetOverallIndex (required), +1 more.
move_view_to_sectionAIdempotent
Move a view (or a section itself) within the sidebar. The single endpoint covers four user actions depending on the arguments:
viewId + sectionId → put the view INTO that section at targetIndex
viewId + sectionId: null → move the view OUT to ungrouped at table-level targetIndex
sectionId-as-viewIdOrSectionId + targetIndex → reorder the section among other sections
viewId + same section → reorder the view within its current section For section reorders, targetIndex is into the table's top-level mixed viewOrder; for in-section moves, it's into that section's viewOrder.
Inputs: appId (required), tableId (required), targetIndex (required), viewIdOrSectionId (required), +2 more.
move_visible_columnsAIdempotent
Move columns by visible-only index (index 0 = leftmost shown column, hidden columns not counted). Use when you want to position relative to what the user sees. Use move_overall_columns when you need to position relative to the full underlying column order including hidden fields. ⚠️ The API preserves existing relative order of supplied IDs — to place columns in a custom sequence, issue one call per column with incrementing targets.
Inputs: appId (required), viewId (required), columnIds (required), targetVisibleIndex (required), +1 more.
query_recordsARead-onlyIdempotent
Read records from an Airtable table view. Returns resolved field values including lookup fields. Supports optional client-side text search across all field values — unlike the REST API filterByFormula approach, this search works correctly on lookup fields. Fetch up to 1000 records per call.
Inputs: appId (required), viewId (required), tableId (required), debug, +3 more.
remove_extensionBDestructive
Remove an installed extension from a dashboard.
Inputs: appId (required), installationId (required), debug.
rename_extensionCIdempotent
Rename an installed extension.
Inputs: name (required), appId (required), installationId (required), debug.
rename_fieldAIdempotent
Rename a field (column) in an Airtable table. Pre-validates the field exists before mutating.
Inputs: appId (required), fieldId (required), newName (required), debug.
rename_record_templateBIdempotent
Rename an existing record template.
Inputs: name (required), appId (required), templateId (required), debug.
rename_tableAIdempotent
Rename a table in an Airtable base.
Inputs: appId (required), newName (required), tableId (required), debug.
rename_viewCIdempotent
Rename a view.
Inputs: appId (required), viewId (required), newName (required), debug.
rename_view_sectionBIdempotent
Rename a sidebar section.
Inputs: name (required), appId (required), sectionId (required), debug.
reorder_view_fieldsAIdempotent
Reorder the fields (columns) displayed in a view. Accepts a partial map: pass only the field IDs you want to move, e.g. { "fldX": 1 } to move fldX to position 1. Other fields keep their relative order. Index 0 is the leftmost position after the primary field. Internally the tool reads the view's current columnOrder, applies the moves, and sends the complete map (the underlying internal API rejects single-key inputs with FAILED_STATE_CHECK — user report 2026-04-30 §2.6).
Inputs: appId (required), viewId (required), fieldOrder (required), debug.
set_calendar_date_columnsAIdempotent
Set the date-column ranges shown on a Calendar view. Each entry is either { startColumnId } for single-point events or { startColumnId, endColumnId } for range events. The array form lets a single calendar overlay multiple date series at once (e.g. "Created date" + "Start → End range" together).
Inputs: appId (required), viewId (required), dateColumnRanges (required), debug.
set_form_metadataAIdempotent
Update one or more legacy-form-view metadata properties in a single call. Unset properties are not touched. Each property fans out to its own atomic Airtable endpoint.
Supported properties: description — intro text shown above the form afterSubmitMessage — "thank you" text after submission redirectUrl — URL to redirect to after submit refreshAfterSubmit — post-submit behavior (e.g. "REFRESH_BUTTON") shouldAllowRequestCopyOfResponse — boolean: show "send me a copy" toggle to respondents shouldAttributeResponses — boolean: track which user submitted (for signed-in respondents) isAirtableBrandingRemoved — boolean: hide Airtable branding (paid plans only)
Note: "form title" is the view name itself — use rename_view to change it. "Field labels on the form" use a per-field endpoint that has not been captured yet.
Inputs: appId (required), viewId (required), debug, description, +6 more.
set_form_submission_notificationAIdempotent
Toggle email-on-submit notifications for a specific user on a form view. Per-user, not per-form (separate from set_form_metadata).
Inputs: appId (required), userId (required), viewId (required), shouldEnable (required), +1 more.
set_record_template_cellAIdempotent
Pre-fill a field value on a record template.
CELL OBJECT TYPES (verified via API capture 2026-05-01):
Static value (text, number, boolean, single-select choice ID): { "type": "static", "value": "some text" } { "type": "static", "value": 42 } { "type": "static", "value": true } { "type": "static", "value": "selXXXXXXXXXXXXXX" } ← single-select: pass choice ID
Linked record(s): { "type": "linkedRows", "value": [{ "foreignRowId": "recXXX", "foreignRowDisplayName": "Record Name" }] }
To clear a field, omit the cellObject or pass null value.
Inputs: appId (required), columnId (required), cellObject (required), templateId (required), +1 more.
set_record_template_visible_columnsAIdempotent
Set which columns are shown (pre-fillable) on a record template. Pass an empty array to show all columns. isPartialSelection:true means only listed columns are shown; false means all are shown.
Inputs: appId (required), columnIds (required), templateId (required), debug, +1 more.
set_view_cell_wrapAIdempotent
Toggle whether long cell values wrap (multi-line) or truncate (single-line with ellipsis).
Inputs: appId (required), viewId (required), shouldWrapCellValues (required), debug.
set_view_color_configAIdempotent
Apply a color config to a view (Kanban / Gallery / Calendar). Currently supports type: "selectColumn" — card colors are taken from a single-select field's choice colors. Other types (e.g. rule-based coloring) exist in Airtable's UI but their payload shapes have not been fully captured yet — passing an unknown type is forwarded as-is so callers can experiment.
Inputs: appId (required), viewId (required), colorConfig (required), debug.
set_view_columnsAIdempotent
One-shot view-column reset: hides every column then shows only visibleColumnIds in the given left-to-right order, with optional freeze. Use this for fresh view setup or full layout rewrites. Use show_or_hide_view_columns when you only want to toggle specific columns without touching the rest.
Inputs: appId (required), viewId (required), visibleColumnIds (required), debug, +1 more.
set_view_coverAIdempotent
Set the cover-image field and crop/fit mode for Kanban or Gallery views. Pass coverColumnId: null to remove the cover. Either field can be passed independently — the other is left untouched.
Inputs: appId (required), viewId (required), debug, coverFitType, +1 more.
show_or_hide_all_columnsAIdempotent
Show or hide every column in a view in one call. Use when you want a clean all-visible or all-hidden baseline. Use set_view_columns when you want to show a specific subset (it hides all then shows only the listed IDs). Use show_or_hide_view_columns for selective per-column toggles.
Inputs: appId (required), viewId (required), visibility (required), debug.
show_or_hide_view_columnsAIdempotent
Show or hide specific columns in a view without affecting others. Pass field IDs + a visibility flag — every listed ID is set to that state, all other columns are untouched. Use set_view_columns instead when you want to define the full visible set from scratch. Use show_or_hide_all_columns to bulk-toggle every column at once.
Inputs: appId (required), viewId (required), columnIds (required), visibility (required), +1 more.
sync_baseADestructiveIdempotent
Base-to-base schema + record sync. IMPORTANT: BOTH mode="plan" and mode="apply" run as BACKGROUND JOBS — they return {jobId, planId, status:"running"} IMMEDIATELY (jobId === planId), NOT a synchronous result; poll mode="status" with that planId to get progress and the final result. mode="plan" (read-only, does NOT mutate): computes an ordered plan (tables/fields to create/update + orphans + warnings) by comparing source/dest schema; when the job finishes, mode="status" returns the plan digest in planDigest. mode="apply" (requires planId from a prior plan): executes the saved plan against the destination — creates tables, reconciles the primary, creates scalar/link/computed fields (source->dest reference remapping + formula validation), applies non-destructive field updates, then runs the RECORD sync (two-pass cells + links, attachments, view-filter restore); aborts if the destination drifted since the plan. mode="status" (poll a plan OR apply job by its planId): returns { phase: "planning"|"schema"|"records"|"done"|"failed", status, recordsMapped, summary, schemaResult?, recordsResult?, result?, planDigest? }. planDigest is human-only by default ({ human, machineOmitted: true }) — pass verbose:true for the full planDigest.machine. Field-mapping errors, APPLY_LOCKED, and DRIFT surface HERE (phase="failed" or an aborted schemaResult), not as a synchronous error. mode="reconcile" (SYNCHRONOUS): rebuild/repair the record map — existence-prune dead idmap entries, optional natural-key re-match per table. mode="diff" (SYNCHRONOUS): schema digest comparing source and destination WITHOUT saving a plan; returns a diffId and summary; pass detail= to drill into a section of a prior diff.
Inputs: mode (required), destAppId (required), sourceAppId (required), skip, +17 more.
update_extension_stateBIdempotent
Enable or disable an extension installation.
Inputs: appId (required), state (required), installationId (required), debug.
update_field_configAIdempotent
Update the configuration of any field — computed OR non-computed. Works for formula, rollup, lookup, count, singleSelect, multipleSelects, number, date, text, and all other field types.
COMMON typeOptions by fieldType:
formula: { formulaText: "IF({Field}, 1, 0)" } rollup: { relationColumnId: "fldLINK", foreignTableRollupColumnId: "fldTARGET", formulaText: "SUM(values)" } (formulaText is REQUIRED; old keys fieldIdInLinkedTable/recordLinkFieldId auto-translated) lookup: { relationColumnId: "fldLINK", foreignTableRollupColumnId: "fldTARGET" } (old keys fieldIdInLinkedTable/recordLinkFieldId auto-translated) count: { recordLinkFieldId: "fldXXX" } singleSelect: { choices: [{ name: "Option A", color: "blue" }], default: "selXXX" } multipleSelects: { choices: [{ name: "PC", color: "blue" }, { name: "Xbox", color: "cyan" }], default: ["selXXX"] } number: { format: "integer"|"decimal"|"currency"|"percentV2", precision: 2, symbol: "$", negative: false } text / multilineText / checkbox: omit typeOptions entirely — passing {} causes a 422
SELECT CHOICES:
Pass choices as array [{ name, color? }] or object { selXXX: { name, color? } }.
Color names (confirmed): "blue", "cyan", "teal", "green", "yellow", "orange", "red", "pink", "purple", "gray".
"default" = pre-selected value: string ID for singleSelect, array of IDs for multipleSelects.
ADDING TO AN EXISTING SELECT FIELD (merge, not replace): Choices not in the list are DELETED. To add without losing existing choices:
Call get_table_schema — each existing choice has { id, name, color }
Pass the full list: existing entries WITH their id, new entries WITHOUT: { choices: [{ id: "selXXXXXXXXXXXXXX", name: "Existing" }, { name: "New Choice", color: "pink" }] }
REPLACING ALL CHOICES: just pass the new choices without any IDs.
Inputs: appId (required), fieldId (required), fieldType (required), typeOptions (required), +1 more.
update_field_descriptionAIdempotent
Update the description text of a field.
Inputs: appId (required), fieldId (required), description (required), debug.
update_formula_fieldAIdempotent
Update the formula body of an existing formula field — shorthand for update_field_config with type "formula". Automatically preserves existing format/precision typeOptions (e.g. percentV2, precision). Use update_field_config to change the field type or other typeOptions.
Inputs: appId (required), fieldId (required), debug, formulaText, +1 more.
update_frozen_column_countAIdempotent
Set the frozen-column divider position for a grid view. The first N columns from the left are frozen and stay visible during horizontal scroll.
Inputs: appId (required), viewId (required), frozenColumnCount (required), debug.
update_recordsA
Update primitive / single-select cells of existing records via cellValuesByColumnId. (Array cells — multi-select, links, attachments — are not set here.) Per-row isolation.
Inputs: appId (required), tableId (required), updates (required), debug.
update_record_template_descriptionAIdempotent
Set or update the description text of a record template.
Inputs: appId (required), templateId (required), description (required), debug.
update_view_descriptionAIdempotent
Update the description text of a view.
Inputs: appId (required), viewId (required), description (required), debug.
update_view_filtersAIdempotent
Update the filter configuration of a view. Supports AND/OR conjunctions, nested filter groups, and Airtable's internal filter operators.
FILTER FORMAT: Leaf filter: { columnId: "fldXXX", operator: "", value: } Nested group: { type: "nested", conjunction: "and"|"or", filterSet: [...] } Clear filters: { filterSet: [], conjunction: "and" } (or pass filters: null)
Filter IDs (flt-prefixed) are auto-generated — do NOT include them.
OPERATORS by field type — verified against Airtable's internal API (2026-04-17 capture; user report 2026-04-30):
Text / URL / Email / Phone:
"=" (exact match — value: string)
"!=" (not equal)
"contains" (value: string)
"doesNotContain"
"isEmpty" / "isNotEmpty" — input-side; auto-rewritten to "=" / "!=" "" before sending (the internal API rejects them on text fields with FAILED_STATE_CHECK)
Number / Percent / Currency:
"=", "!=", "<", ">", "<=", ">=", "isEmpty", "isNotEmpty"
Single select:
"=" (value: "selXXX" — the choice ID, NOT the choice name)
"!="
"isAnyOf" / "isNoneOf" (value: ["selXXX", "selYYY"] — array of choice IDs)
"isEmpty" / "isNotEmpty"
Multiple select:
"hasAnyOf", "hasAllOf", "hasNoneOf", "isExactly", "isEmpty", "isNotEmpty"
Checkbox:
"=" (value: true|false)
Date (absolute):
"is", "isBefore", "isAfter", "isOnOrBefore", "isOnOrAfter", "isEmpty", "isNotEmpty"
value: ISO date string e.g. "2026-01-15"
Date (relative) — "isWithin":
value: { "mode": "", "timeZone": "", "shouldUseCorrectTimeZoneForFormulaicColumn": true }
timeZone: IANA string e.g. "Europe/Istanbul", "America/New_York", "UTC"
Modes (no numberOfDays): "pastWeek", "pastMonth", "pastYear",
"nextWeek", "nextMonth", "nextYear",
"thisCalendarMonth", "thisCalendarYear"
Modes (add numberOfDays key): "pastNumberOfDays", "nextNumberOfDays"
Example — past week: { "operator": "isWithin", "value": { "mode": "pastWeek", "timeZone": "UTC", "shouldUseCorrectTimeZoneForFormulaicColumn": true } }
Example — past N days: { "operator": "isWithin", "value": { "mode": "pastNumberOfDays", "numberOfDays": 7, "timeZone": "UTC", "shouldUseCorrectTimeZoneForFormulaicColumn": true } }
Example — this month: { "operator": "isWithin", "value": { "mode": "thisCalendarMonth", "timeZone": "UTC", "shouldUseCorrectTimeZoneForFormulaicColumn": true } }
Formula / Lookup / Rollup (text result type):
Same as Text. "isEmpty" / "isNotEmpty" are auto-rewritten to "=" / "!=" "".
Linked record (foreignKey):
"contains" (value: linked record name) works.
"isEmpty" / "isNotEmpty" do NOT work — the call throws a clear error directing
you to a helper formula like IF(LEN({Linked} & "")>0,"yes","") and a "=" / "!=" filter on that helper.
AUTO-NORMALIZATION (applied client-side before the request):
"is" → "=" (the internal API does not recognize "is")
"isNot" → "!="
"isAnyOf" with a single-element array or scalar value → "=" with scalar value
"isEmpty" → "=" "" on text / formula(text) / lookup(text) / rollup(text) fields
"isNotEmpty" → "!=" "" on text / formula(text) / lookup(text) / rollup(text) fields For single-select, value must be the choice ID (selXXX) — use get_base_schema to find IDs.
NESTING LIMIT:
The internal API accepts at most 2 levels of nesting (top conjunction + one
layer of nested groups). Deeper trees are rejected with FAILED_STATE_CHECK.
Workaround: flatten by repeating shared conditions inside each leaf group,
e.g. (A AND B) OR (A AND C) instead of A AND (B OR C) if you need
another nested AND inside the OR. The error message returned by this tool
flags depth-related failures explicitly.
EXAMPLES: Text equals: { filterSet: [{ columnId: "fldXXX", operator: "=", value: "Prime" }], conjunction: "and" } SingleSelect equals: { filterSet: [{ columnId: "fldXXX", operator: "=", value: "selABC123" }], conjunction: "and" } Text contains: { filterSet: [{ columnId: "fldXXX", operator: "contains", value: "hello" }], conjunction: "and" } Number range: { filterSet: [{ columnId: "fldX", operator: ">=", value: 10 }, { columnId: "fldX", operator: "<=", value: 100 }], conjunction: "and" } Nested (a AND (b OR c)): { filterSet: [{ columnId: "fldA", operator: "contains", value: "x" }, { type: "nested", conjunction: "or", filterSet: [{ columnId: "fldB", operator: "=", value: 1 }] }], conjunction: "and" }
Inputs: appId (required), viewId (required), filters (required), debug, +1 more.
update_view_group_levelsAIdempotent
Set grouping on a view. Default mode replaces all existing group levels — pass an empty array with operation="replace" to clear grouping. Use operation="append" to add new group levels below the existing ones without rewriting them.
Inputs: appId (required), viewId (required), groupLevels (required), debug, +1 more.
update_view_row_heightAIdempotent
Change the row height of a grid view.
Inputs: appId (required), viewId (required), rowHeight (required), debug.
upload_attachmentA
Upload attachments into an attachment cell by URL. Airtable's servers fetch each URL directly (the UI's 'Add attachment → Add URL') — bytes are NOT proxied through this server. Use for multipleAttachments fields, which update_records cannot set. Appends to the cell (calling twice adds two). Per-update isolation: a failing update is reported, not fatal.
Inputs: appId (required), updates (required), debug.
validate_formulaARead-onlyIdempotent
Validate a formula expression before creating or updating a formula field. Returns whether the formula is valid and what result type it produces (text, number, etc). Use this before create/update to catch errors early.
Inputs: appId (required), tableId (required), formulaText (required), debug.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 73 tools
There are many overlapping clusters: list_tables/get_base_schema/get_table_schema/list_fields, create_field/create_formula_field/update_field_config/update_formula_field, and a large family of view-column tools (set_view_columns, show_or_hide_view_columns, show_or_hide_all_columns, move_visible_columns, move_overall_columns, reorder_view_fields). Detailed descriptions help, but several tools have genuinely unclear boundaries and are easy to misselect.
Names are consistently snake_case and mostly follow a verb_noun pattern. There are minor deviations with compound names like show_or_hide_view_columns, create_extension_dashboard, set_view_cell_wrap, and similar actions using different verbs (set/update/apply), but the convention remains predictable.
At 73 tools, the surface is far beyond a well-scoped set and includes many micro-operations for views, record templates, extensions, daemon management, and sync. The manage_tools profile mechanism mitigates this somewhat, but the default full profile is still extremely heavy for an agent to navigate.
Coverage is broad across tables, fields, views, records, record templates, formulas, extensions, view sections, and base-to-base sync, with CRUD for most core resources. Minor gaps remain around base-level lifecycle, comments/webhooks/automations, and some record cell types handled indirectly through separate tools.
Maintenance
Related MCP Connectors
Official Airtable MCP server — database and operations layer for agents.
Unofficial Airtable MCP Pack — wraps the Airtable REST API v0
Hosted Airtable MCP server for Claude, ChatGPT and Cursor, with each action switched per account.
Glide's official MCP server — build, manage, and operate GlideOS apps, data, and workflows.
Related MCP Servers
- AlicenseBqualityBmaintenanceA Model Context Protocol server that provides read and write access to Airtable databases. This server enables LLMs to inspect database schemas, then read and write records.163,650 npm456TypeScriptMIT
- AlicenseAqualityDmaintenanceA Model Context Protocol server that provides tools for programmatically managing Airtable bases, tables, fields, and records through Claude Desktop or other MCP clients.12291 npm75MIT
- AlicenseNot gradedqualityDmaintenanceA production-ready MCP server for Airtable that enables programmatic management of bases, tables, fields, and records through Claude Desktop or other MCP clients using Effect for type-safe and robust API interactions.6 npm1MIT
- AlicenseNot gradedqualityDmaintenanceExtends Airtable MCP server with full CRUD, bulk import/upsert, data analysis, schema validation, and data export capabilities.MIT