Connect AI Gmail

CData Connect AI — Gmail Skill

⚠️ Prerequisites — load these first

• connect-ai-base skill

Before proceeding, locate and read the connect-ai-base skill. If it is not available in the current environment (not loaded, not enabled, or not installed), stop immediately. Do not proceed with the task. Tell the user that the connect-ai-base skill is required and ask them to install and enable it before continuing.

This skill provides Gmail-specific guidance for querying Gmail data through CData Connect AI. It composes on top of the connect-ai-base skill, which handles the discovery workflow, error recovery, SQL dialect, and query naming convention.

Precedence

This skill replaces getInstructions for the Gmail driver. Do not call getInstructions for Gmail — the guidance it provides is already incorporated here. Proceed directly to schema discovery (getTables / getColumns) after identifying the Gmail connection via getCatalogs.

Schema

Gmail is a single-schema driver. The schema name is REST. (IMAP is a legacy schema that can be selected through connection properties; the REST schema is the default and is assumed throughout this skill.)

Three-part names take the form:

SELECT [Subject], [From], [Date]
FROM [YourConnection].[REST].[Messages]
WHERE [SearchQuery] = 'is:unread'
ORDER BY [Date] DESC
LIMIT 10

Replace YourConnection with the actual Gmail connection (catalog) name from getCatalogs.

Query Process

1. Identify the object you need. Most requests target Messages (or a label view), Threads, Labels, or Attachments. Use getColumns on the target before querying.
2. Filter messages with SearchQuery. For the Messages table, the SearchQuery pseudo-column is the most efficient filter — it passes native Gmail search operators to the server and takes precedence over any other SQL WHERE criteria. Fall back to ordinary column predicates only for fields Gmail search does not cover.
3. Fetch a single message for full detail. Some columns (AttachmentIds) and the Labels count columns are only populated when you filter to a single row by Id. If a bulk list shows those columns empty, re-query the one row by Id.
4. Retrieve attachment content by filtering the Attachments table on MessageId, or by calling the DownloadAttachments procedure — both return base64 in cloud environments (see Stored Procedures).

Data Model

Key Tables and Views

• Messages — Core email data (subject, sender, recipients, date, snippet, body, labels, attachment IDs). The central table for most queries.
• Draft — Unsent draft messages; the same email fields as Messages plus a MessageId. Supports INSERT to create a draft.
• Sent — Sent messages.
• Threads — Email conversation threads, keyed by Id, with a Snippet and HistoryID.
• Labels — Gmail labels: display Name, Type (system or user), and message/thread counts.
• MessageLabels — Join table with one row per (message Id, LabelId) pair. Use it to find which labels a message carries, or which messages carry a label, without parsing the comma-separated Labels string on Messages.
• Attachments — Attachment metadata and base64 content, keyed by MessageId + Id.
• Settings tables — AutoForwarding, Delegates, Filters, ForwardingAddresses, Language, SendAs, SendAsAliasSmimeInfo, Vacation, Users. Query these for account configuration rather than mail content.

The label-as-view pattern

Every Gmail label — system and user-created — surfaces as its own view named after the label. Each view is a pre-filtered set of messages that shares the Messages columns.

• System label views: INBOX, UNREAD, STARRED, IMPORTANT, SENT, SPAM, TRASH, CHAT, and the categories CATEGORY_PERSONAL, CATEGORY_SOCIAL, CATEGORY_PROMOTIONS, CATEGORY_UPDATES, CATEGORY_FORUMS.
• User label views: one per user-created label (e.g. Work, Receipts, and any nested Parent/Child labels).

Because of this, a full getTables listing is long and volatile — it grows and changes as the account's labels change, and includes many ad-hoc user labels. Do not rely on the full table list being stable. To filter by a well-known set, query the matching system view (e.g. UNREAD); to discover labels programmatically, query the Labels table; to filter Messages by arbitrary label IDs, use the LabelsFilter pseudo-column.

Key Relationships

• Messages.ThreadId → Threads.Id — messages belong to a thread.
• Messages.Id → MessageLabels.Id, and MessageLabels.LabelId → Labels.Id — the many-to-many message↔label relationship.
• Attachments.MessageId → Messages.Id — attachments belong to a message; Messages.AttachmentIds lists the IDs.

Important Columns

Messages

• Id — Unique message ID (primary key).
• Subject, From, To, CC, BCC — Subject and addresses.
• Content — Full message body.
• Snippet — Short preview of the body.
• Date — Timestamp the email was sent.
• Size — Size of the email in bytes.
• Labels — Comma-separated list of label IDs applied to the message.
• AttachmentIds — Comma-separated attachment IDs; populated only when fetching a single message by Id.
• AttachmentPath — Path used when supplying attachments on write.
• ThreadId — ID of the containing thread.
• HistoryId, Headers — History record ID and full header list.
• RawMessage — Full RFC 2822, base64url-encoded message; only returned when MessageFormat=raw.
• SearchQuery (filter-only pseudo-column) — Native Gmail search string; evaluated server-side and takes precedence over other SQL criteria.
• LabelsFilter (pseudo-column) — A comma-separated list of label IDs (just the list — there is no AND operator in the syntax). Results contain only messages carrying all listed labels (e.g. 'UNREAD,IMPORTANT' returns messages that are both unread and important).
• IncludeSpamTrash (pseudo-column) — Set to true to include SPAM and TRASH messages, which are excluded by default.
• MessageFormat (pseudo-column) — minimal, full, raw, or metadata (default full).

Labels

• Id — Immutable label ID (e.g. INBOX, UNREAD, or a user-label ID). Primary key, read-only.
• Name — Display name (writable).
• Type — system or user (read-only).
• MessageListVisibility — SHOW or HIDE.
• LabelListVisibility — LabelShow, LabelHide, or LabelShowIfUnread.
• MessagesTotal, MessagesUnread, ThreadsTotal, ThreadsUnread — Counts; populated only when filtering to a single label by Id, not in bulk list queries.

Attachments

• Id — Attachment identifier within the message (returned as a sequential index, e.g. 1, 2, 3).
• MessageId — ID of the containing message. Required as a filter — the table cannot be queried without it.
• Filename — Attachment file name.
• Size — Size in bytes. Not populated in metadata-only mode (IncludeAttachmentData = false); returned only on a full data pull or via DownloadAttachments.
• Data — Attachment content encoded as URL-safe base64 (uses - and _ instead of + and /). Decode with a URL-safe base64 decoder (e.g. Python base64.urlsafe_b64decode), not a standard base64 decoder.
• IncludeAttachmentData (pseudo-column) — Whether to include Data (default true); set to false to fetch metadata only. In metadata-only mode Size comes back empty — a full data pull (or DownloadAttachments) is required to get Size.

MessageLabels

• Id — Message ID (key).
• LabelId — Label ID applied to that message (key). One row per applied label.

Common Query Patterns

Search messages with Gmail search syntax

SearchQuery is the most efficient filter for Messages and takes precedence over other SQL criteria.

SELECT [Subject], [From], [Date]
FROM [YourConnection].[REST].[Messages]
WHERE [SearchQuery] = 'has:attachment from:google.com newer_than:30d'
ORDER BY [Date] DESC
LIMIT 10

Recent unread messages

Gmail has no isRead column — read state is a label — so query the UNREAD view (or is:unread in SearchQuery). The view returns newest-first natively, so LIMIT 10 alone is enough — do not add ORDER BY [Date] here (sorting the full UNREAD set times out; see Gmail-Specific Conventions).

SELECT [Subject], [From], [Date], [Snippet]
FROM [YourConnection].[REST].[UNREAD]
LIMIT 10

List labels

Count columns are not populated when listing all labels.

SELECT [Id], [Name], [Type]
FROM [YourConnection].[REST].[Labels]
ORDER BY [Name]

Get message and thread counts for one label

Filter to a single label by Id to populate the count columns.

SELECT [Id], [Name], [MessagesTotal], [MessagesUnread], [ThreadsTotal]
FROM [YourConnection].[REST].[Labels]
WHERE [Id] = 'INBOX'

Restrict messages to specific labels

Use LabelsFilter with a comma-separated list of label IDs — just the list, no AND operator in the syntax. Results contain only messages carrying all listed labels (below: messages that are both unread and important).

SELECT [Subject], [From], [Date]
FROM [YourConnection].[REST].[Messages]
WHERE [LabelsFilter] = 'UNREAD,IMPORTANT'

Retrieve attachment content

Query the Attachments table filtered by MessageId. Data is returned as base64. (The DownloadAttachments procedure returns the same content — see Stored Procedures.)

SELECT [Id], [Filename], [Size], [Data]
FROM [YourConnection].[REST].[Attachments]
WHERE [MessageId] = '19f1c11dc1070717'

To list attachment metadata without pulling the (potentially large) Data payload, set IncludeAttachmentData to false. Note Size is not populated in this mode, so omit it:

SELECT [Id], [Filename]
FROM [YourConnection].[REST].[Attachments]
WHERE [MessageId] = '19f1c11dc1070717' AND [IncludeAttachmentData] = false

Stored Procedures

SendMailMessage

Sends an email. Parameters: To, Subject, Content (body), and optional From, CC, BCC, Attachments (a temp-table name or JSON aggregate of attachment content). Always confirm with the user before sending.

EXEC [YourConnection].[REST].[SendMailMessage] To = 'recipient@example.com', Subject = 'Status update', Content = 'The report is attached.'

JSON form:

{
  "procedure": "SendMailMessage",
  "parameters": {
    "To": "recipient@example.com",
    "Subject": "Status update",
    "Content": "The report is attached."
  }
}

ReplyToMailMessage

Replies to an existing message, preserving the thread. Supply the message ID being replied to plus the reply body.

SendDraft

Sends an existing draft to the recipients in its To/Cc/Bcc headers. Supply the draft ID.

UpdateMessageLabels

Adds or removes labels on one or more messages (mark read/unread, archive, star, etc.). Parameters: MessageIds (up to 1000, comma-separated), LabelsToAdd, LabelsToRemove (label IDs, up to 100 each). Marking a message read means removing the UNREAD label.

EXEC [YourConnection].[REST].[UpdateMessageLabels] MessageIds = '19f1c11dc1070717', LabelsToRemove = 'UNREAD'

TrashMessage / UntrashMessage

Move a message to or from the trash by message ID. Confirm before trashing.

DownloadAttachments

Downloads all attachments of a single message. Parameters: MessageId (required), and optional AttachmentId and FileStream. Do not pass FileStream — it is an output-stream object that the Connect AI interface cannot supply (and there is no DownloadLocation disk-path parameter on this driver). Calling with just MessageId (optionally AttachmentId for a single attachment) does return a row per attachment with columns Success, MessageId, AttachmentId, Size, Data (base64), and Filename.

{
  "procedure": "DownloadAttachments",
  "parameters": { "MessageId": "19f1c11dc1070717" }
}

This is the procedure equivalent of querying the Attachments table — the two are interchangeable, not one a workaround for the other. Use whichever fits: DownloadAttachments returns every attachment for a message in one call; the Attachments table lets you select individual columns and skip the payload via IncludeAttachmentData = false.

Settings and account procedures

GetUserProfile, GetAutoForwarding / UpdateAutoForwarding, GetVacations / UpdateVacations, GetLanguage / UpdateLanguage, GetImap / UpdateImap, GetPop / UpdatePop, ImportMessage, StartNotifications / StopNotifications, and the send-as S/MIME procedures (SetDefaultSendAsAliasSmimeConfig, VerifySendAs). Use these for account configuration and mailbox administration.

Write Operations

Gmail supports writes through both stored procedures (sending, replying, labeling, trashing — above) and direct DML on writable tables:

• Create a draft — INSERT into Draft (To, Subject, Content, and optionally From, CC, BCC are writable).
• Create or rename a label — INSERT or UPDATE Labels (Name, MessageListVisibility, LabelListVisibility are writable).
• Forwarding, vacation, delegates, send-as — the corresponding settings tables/procedures.

Write access is governed by the Connect AI catalog permissions for the Gmail connection. Check the PERMISSIONS column from getCatalogs: a connection must show Insert / Update / Delete / Execute (not just Select) for the corresponding operation to succeed. If a write is rejected, the connection is read-only for the current user and access must be granted in Connect AI.

No DELETE path over MCP. The Connect AI MCP surface exposes no delete operation (there is no execute_delete), even when the connection has Delete permission — so a row cannot be removed with DELETE FROM ... over MCP. To discard a draft, call the TrashMessage procedure with the draft's MessageId (from the Draft table's MessageId column).

Gmail-Specific Conventions

• Every label is a view. The table list contains one view per Gmail label (system and user), so it is long and changes with the account. Discover labels via the Labels table; filter via system views (UNREAD, INBOX, …) or the LabelsFilter pseudo-column rather than trusting the raw table list.
• SearchQuery beats SQL predicates. It runs server-side and takes precedence over other WHERE criteria. Use Gmail operators (from:, to:, subject:, has:attachment, after:, before:, newer_than:, is:unread, label:) for the fastest, most accurate filtering.
• has:attachment over-matches. It also matches delivery-status-notification and message/rfc822 messages whose AttachmentIds is empty (no downloadable file parts). Treat an empty AttachmentIds as "no files to download" even when has:attachment matched the message.
• Avoid ORDER BY [Date] on large or unfiltered sets. Sorting a big result set (e.g. all UNREAD, ~20k rows) times out and drops the MCP connection. It is fine over a SearchQuery-filtered small set. Prefer narrowing with SearchQuery before sorting, or rely on Gmail's native newest-first order.
• No read/unread column. Read state is the UNREAD label. Query the UNREAD view or is:unread; mark read by removing the UNREAD label via UpdateMessageLabels.
• Single-fetch-only columns. AttachmentIds on Messages and the count columns on Labels populate only when you query one row by its Id, not in bulk list queries.
• Attachments require a MessageId filter and return content as base64 in Data. Set IncludeAttachmentData = false to skip the payload when you only need metadata.
• SPAM and TRASH are excluded by default. Set IncludeSpamTrash = true to include them.
• Confirm before any mail action — sending, replying, trashing, or relabeling changes the user's live mailbox.