The MCP server is in public beta. It is open to use and still being refined, so tools and behavior may change.
Tool
get_vault_documents · Behavior Read-onlyfilters and sort rather than paging everything. values is keyed by column field_key; read the keys, labels, and types from get_vault_columns first. A value is a string, and structured columns carry JSON in it: a currency holds {"amount","currency"}, a date holds YYYY-MM-DD, a multi-select holds a JSON array. null means the scan found no value. A document appears here once its first scan writes values, a few minutes after upload, so a row missing right after upload_file is not lost. To answer from the text inside the documents rather than the extracted columns, use ask_gcai with this vault instead.
Input parameters
string (uuid)
required
The vault to read, from
get_vaults.string (uuid)
Fetch one row by its
document_id. Provide either document_id or a list query (q/filters/sort/cursor), not both.string
Case-insensitive substring of the document name. This searches names only, not the text inside the documents. To answer a question from the contents, use
ask_gcai with this vault_id instead.object[]
Rules combined with AND. A rule naming an unknown column, or an operator that column cannot use, is an error rather than a rule that is quietly dropped, so a count is never silently wrong.
object
Sort by a column key, or by
fileName, extractedAt, or sourceName.string
Opaque pagination cursor from a previous response’s
next_cursor. Omit for the first page; repeat any filters (e.g. q) alongside it.integer
Max items to return (default 25, max 100). Defaults to
25.Response
object[]
number | null
string | null