Added an official MCP server for connecting Claude Code, Codex, Cursor, and other compatible AI clients to Teable through guided OAuth authorization, without manually creating a personal access token. Self-hosted instances support connections at /mcp without additional configuration.
Connected AI clients can list projects, inspect tables and fields, query and manage records, and discover APIs for advanced operations, all within the user’s existing permissions.
Added Settings → MCP to review and revoke AI client connections. Personal access tokens remain available for CI and other browserless environments.
Bug Fixes & Improvements
Fixed an issue where tables could remain unavailable after a network timeout. Once connectivity is restored, retrying reloads the same table without requiring users to reopen it from the project directory.
Single-line text fields now trim leading and trailing whitespace and replace line breaks and tabs with spaces when records are created or updated. Integrations that rely on preserving this whitespace should account for the change. Long-text fields continue to preserve line breaks.
Added This Year, Last Year, and Next Year presets to date filters.
Added “Add to chat” to the record detail menu, with correct context for records opened from linked tables or outside the current filtered view. On mobile, the action opens AI chat and reliably adds the record.
Skill activation now appears as a dedicated step in web and mobile chats, showing the skill name, description, and expandable instructions. Messages starting with /skill-name now reliably activate available skills.
Large web fetch and tool results are now available as readable files, with clearer failure handling for tool actions.
Bug Fixes & Improvements
Improved platform security and reliability.
Creating a user field through the API without specifying shouldNotify now leaves notifications off. Explicitly setting it to true continues to send notifications.
AI-generated list views now display all stored numeric priorities, including values outside the standard High, Medium, and Low mapping.
App Builder sessions now consistently apply the latest field-display rules, including retained sessions and existing workspaces.
New single-line text values now trim surrounding spaces and replace line breaks and tabs with spaces, including writes without typecasting. Pasting multiline values into single-line primary fields also reuses matching linked records more reliably.
Long text values now trim leading and trailing whitespace while preserving internal line breaks and tabs. Values that are empty after trimming are stored as null.
Required form fields without default values now appear even if previously hidden. Required fields with defaults remain hidden as configured.
URLs in single-line and long-text cells now open on the first click, even when the cell is not selected.
Rating and checkbox icons now apply values on the first click. Clicking the current rating or a checked checkbox again clears it without an extra click.
Percentage fields now display and save values consistently across forms, expanded records, and grids. For example, entering 50 saves 0.5 and displays 50%.
Voice input now uses OpenAI’s gpt-transcribe model to maintain transcription availability. Saved configurations using legacy transcription models are automatically mapped to the new model.
Restored embedded child tables for accessible linked records, including headers, values, and record actions. Read-only record details also preserve this view and allow linked records to open; share views continue to use compact tags.
Text-to-number conversion now correctly handles currency values, numbers within text, and percentages. Text without a supported numeric value still converts to an empty value.
Number-to-checkbox conversion now treats all non-empty numeric values, including 0, as checked. Only empty values remain unchecked.
Saving single-select fields with duplicate option names now shows a clear validation error and leaves the field unchanged. Duplicate checks also ignore surrounding spaces.
Multiple-select to single-line text conversion now quotes option names containing commas, double quotes, or line breaks to preserve value boundaries.
Single- and multiple-select fields now correctly parse quoted typecast values containing commas. Select writes consistently use option names: non-typecast requests reject option IDs, while typecast requests treat ID-like values as literal names.
Table creation now returns VALIDATION_ERROR instead of validation_error when an unsupported field type is used as the sole primary field. These requests continue to be rejected without creating a table.
Date writes through the v2 API now respect field date formats and time zones, with or without typecasting. Date-to-text conversion preserves the displayed format and time zone, while text-to-date conversion uses the target field’s time zone and clears invalid values.
Renaming or deleting select options now updates unchanged defaults appropriately, preventing invalid defaults from blocking field saves or record creation. Explicit default changes remain respected, and resubmitting unchanged options without IDs no longer clears valid defaults.
Restored the time picker for empty date cells configured with a 24-hour format. Date-only inputs and calendar selections now save at midnight in the field’s configured time zone.
Formulas referencing button fields now return empty values instead of button-count data. Rollups over empty linked-record sets now return empty values for average and xor, so some existing results may change from 0 or false to empty.
Fixed cross-chat card dismissals so activity in one chat no longer closes pending cards in another. Also eliminated duplicate storage of presented files.
Editors can now import CSV files and append data to existing tables without owner-level permissions. Imported records identify the importing editor as their creator.
Added reusable structured queries with field selection, aliases, filters, sorting, pagination, and aggregations, available through a unified API endpoint and CLI.
Added a versioned Query Library with stable query IDs, immutable revisions, pinned or latest-version references, sharing controls, revision restoration, and protection against conflicting updates.
Bug Fixes & Improvements
Field PATCH requests now correctly apply supported options and constraints, including select choices, lookup display settings, rollup and formula configuration, and unique and required settings. Dependent fields and calculated values update accordingly while preserving applicable formatting. Unsupported options and constraints now return validation errors instead of success without applying changes.
Last modified time and user fields now explicitly represent “track all” with an empty field list, remove references to deleted fields, and preserve existing history during configuration changes.
Assignees now receive member-field notifications when records are duplicated.
Converting fields to single-line text now produces readable filename (token) entries for attachments, clears button click-count data, and quotes user names containing commas.
Conditional rollups now return a clear validation error for source fields outside the selected target table and prevent invalid fields from being created.
Improved reliability for automation workflows, AI field generation, workflow AI actions, routines, and media generation, reducing failures before execution.
Updated Simplified Chinese filter labels and ordering for single-user and multi-user fields.
Restored the ability for base owners and permission matrix administrators to view, restore, and permanently delete archived records when the authority matrix is enabled.
The AI chat model picker is now scrollable, with effort settings remaining visible while users browse models.
Button click limits now require positive integers in both the UI and API. Invalid API updates leave existing settings unchanged. Disabling a limit clears its maximum and reset settings, while unsaved edits are preserved without being overwritten by external field updates.
Added a cookie consent banner for app visitors in the EU, UK, and Switzerland when needed, with consent choices shared with the Teable website. The sign-in page now includes Cookie settings and separates product update email preferences from legal consent.
Bug Fixes & Improvements
AI conversations are now created only when the first message is sent and appear in history after the first successful message, reducing empty chats and keeping conversations accessible without refreshing.
Improved AI chat and App Builder history reliability: unread indicators recover after reconnecting, and conversation actions handle missing or newly created chats more consistently.
Fixed computed-field error handling after linked-record fields are deleted or restored. Dependent formulas now show invalid states correctly, and lookup and conditional rollup values recover more reliably after undo.
Fixed numeric-to-date field conversion failures when dependent formulas or rollups exist. Dependent fields now update more reliably after type changes and clear incompatible formatting and display options.
Fixed built-in email client unsubscribe buttons for automation emails, Send Email API messages, and newsletters, including previously delivered emails with unsubscribe support. The existing confirmation page remains unchanged.
Fixed misleading success messages when pasted link-field values cannot be created or linked. Unsupported input is now validated more consistently, with clearer errors when linked records cannot be created.
Fixed formula field updates through the field PATCH API that could report success without applying the new expression. Valid formula options are now preserved, and result type changes apply more consistently.
Fixed conditional rollup SUM calculations to include all matching records, regardless of sorting or row limits, preventing undercounted totals.
Fixed multi-value lookups inheriting unsupported ring display settings. Compatible number formatting, such as currency and precision, is still preserved.
Lookup and conditional lookup API requests now reject button fields as source fields. Requests that previously accepted these unsupported configurations will now fail validation.
Fixed default naming and function selection for conditional rollups created through the API. Unnamed fields use source-field and target-table names, and an omitted expression defaults to counting all records. Formatting options now appear only after a rollup function is selected.
Fixed single-value formula cells using ERROR("custom error") appearing blank. They now display #ERROR: custom error and retain error details when saved or referenced.
Conditional lookups now return all matching records when the match limit is left blank, rather than defaulting to one match. The configuration hint also reflects the current limit. Existing configurations with a blank limit may now return more records.
Fixed button click counts not being saved through record create, update, and batch update APIs. Saved counts now appear consistently in API responses and subsequent reads.
Fixed TODAY() to respect the formula field’s configured time zone and return the expected local start of day. Time zone handling is also more consistent in nested formulas.
Improved the Simplified Chinese circular dependency message for conditional rollup filters, replacing raw English text and technical identifiers with a localized prompt. Invalid updates remain blocked without changing the existing configuration.
Added a scraping data source catalog to discover and run endpoints by platform and data type, with data source information displayed in the menu. Agent and CLI workflows now support searching, viewing, and running endpoints and checking endpoint status, with clearer data source and endpoint details.
Scraping charges are now calculated based on the actual cost of each data source, with support for reserving credits in advance, setting a per-run credit limit, and settling charges at the end.
Removed Base-level read-only database connections and the associated connection panel. Base no longer provides credentials for direct PostgreSQL access; existing BYODB and SQL query functionality is unaffected.
Bug Fixes & Improvements
Record updates containing only system or computed fields now succeed without changing existing values. Ignored fields such as Auto Number and Created Time remain ignored, while unknown fields still return validation errors.
Scraping menus on web and mobile now display specific data types, such as Posts, Profile, or Company, only when those types are available. Search results and platforms without distinct data types still use platform-only selection.
Improved sandbox security and reliability while retaining AI conversations, app previews, file uploads, and signed preview links.
Fixed failures in imports and bulk record creation caused by emoji near the truncation limit in long text, including Google Sheets imports. Also improved the reliability of text containing special Unicode characters during import rollback.
Project viewers can now find and open completed routine runs and the conversations they generate. Read-only access remains separate from editing actions, and permission consistency across routine history views has been improved.
Fixed an issue where forms blocked submission when required fields with configured default values were left blank. These fields now use their default values.
Improved member avatar portability across deployments and fixed delayed avatar updates in member, creator, and modifier views.
Fixed horizontal scrollbar synchronization issues when dragging across many views, including when dragging to the left and right edges.