DocQuery
Read-only MCP over your Excel or Airtable plus the PDFs beside it, joined as one relational dataset.
我該用這個嗎
品質與安全性
根據工具定義與協定合規性的自動化分析。
上下文成本
這是每次將伺服器的工具載入模型上下文時所消耗的約略 token 數量。數量越高,可用於其他工作的注意力就越少。
安裝
一鍵安裝
將以下內容加入你的 `claude_desktop_config.json` 檔案:
{
"mcpServers": {
"docquery": {
"url": "https://docquery.io/api/v1/mcp"
}
}
}遠端端點
https://docquery.io/api/v1/mcpstreamable-http它能做什麼
工具清單
工具(4)
🟢list_tables
List all tables in the workspace. Call this first to learn table and column slugs. Returns slug, name, row_count, and columns (slug, name, type, primary_key) for each table. Document-extracted tables include source_document {id, filename, url}. source_document is null for CSV, Excel, Google Sheets, and Airtable. url is the signed-in document page; original file bytes are not stored, so it is not a download. Cite source_document.filename when you report a value taken from a document.
輸入結構描述
{
"type": "object",
"properties": {},
"required": []
}🟢list_relations
List the relations (foreign keys) between tables in the workspace. Each relation has from and to as "table_slug.column_slug" and a kind. Use it to find which columns link two tables before calling related_rows. Read-only; returns an empty list when no relations are defined.
輸入結構描述
{
"type": "object",
"properties": {},
"required": []
}🟢query_table(offset, filter, search, columns, limit, ...)
Read rows from a table. Optional filter is an exact-match map of column slug => value; optional search is a case-insensitive substring match across all cells. Returns source_document for the table (null for spreadsheets and Airtable). Each row includes source_page: the 1-based PDF page, PowerPoint slide, or Google Slides slide the row was extracted from, or null when the page is unknown (Word, ODT, RTF, spreadsheets, Airtable, manually created rows, or a row that spans pages). Cite source_document.filename and source_page when an extracted value looks wrong. Do not invent a page when source_page is null. If a column is already named source_page, the page is returned as _source_page.
輸入結構描述
{
"type": "object",
"properties": {
"offset": {
"default": 0,
"type": "integer",
"description": "Rows to skip, for paging past the first page. Use offset + returned while it is below total."
},
"filter": {
"type": "object",
"description": "Optional exact-match filter: column slug => value.",
"additionalProperties": {
"type": "string"
}
},
"search": {
"type": "string",
"description": "Optional case-insensitive text search: keeps rows where any cell contains this text (e.g. part of a name or invoice number). Combines with filter; total counts matches."
},
"columns": {
"type": "array",
"description": "Optional column slugs to return (from list_tables). Omit for all columns. source_page is always included.",
"items": {
"type": "string"
}
},
"limit": {
"default": 25,
"type": "integer",
"description": "Maximum rows to return (default 25, max 100)."
},
"table_slug": {
"type": "string",
"description": "Slug of the table to read (from list_tables)."
}
},
"required": [
"table_slug"
]
}🟢related_rows(value, column, table_slug)
Follow a relation from a source table to the rows it links to. Given a column on `table_slug` that participates in a relation and a `value` (usually a cell from a source row), returns the matching rows in the linked table. Works in both directions. Each linked table includes source_document when it came from an uploaded document, and each row includes source_page (or null). Cite the filename and page; do not invent a page.
輸入結構描述
{
"type": "object",
"properties": {
"value": {
"type": "string",
"description": "The value to match across the relation."
},
"column": {
"type": "string",
"description": "A column on table_slug that participates in a relation."
},
"table_slug": {
"type": "string",
"description": "Slug of the source table (from list_tables)."
}
},
"required": [
"table_slug",
"column",
"value"
]
}社群
證據