カテゴリーを見る

Shape First: Row Count and Size Before Your AI Client Reads a Table

5分で読める

0:00 6:52
Analyse your data with your AI client · Shape first

Know it is 60 million rows before you ask for them

Columns, a row-count estimate, the size of one row and of the whole result, and how many rows fit your budget. No rows are sent.

Personal Core and up Runs on your Network Agent 読み取り専用、常に
行き詰まったら、遠慮なく質問してください。 開く サポート クエリ ストリーム（右上メニュー → サポート）からメッセージを送信してください。担当者が目を通しますので、「この画面で何を求めているのか分かりません」というご質問は、全く問題ありません。

What your Network Agent does for you

Query Streams MCP iconYour AI client learns the size of a result before it pays for it. Your Network Agent reads the first rows, measures them, estimates the total and stops the query, then sends that back with no rows at all. The shape of a 60-million-row table comes back in under a second.

Asking for rows is how an AI client usually finds out a table is huge: the result floods in, gets cut, and the question is asked again. Shape first turns that around. Your AI client asks for the shape, then decides: a few rows, a プロフィール, an aggregate, or a narrower query.

状況BeforeWith Shape first
“What is in the inventory detail table?”A thousand raw rows, cut by the client, size unknown.About 60 million rows, its columns, tokens per row, and the empty columns named.
“Profile it.”Nobody can tell how long that will take.The shape reports how fast the database delivers rows, so the client can size the profile.
A join across two big tablesRuns until it times out.An estimate first, so the client adds a filter.

The plan it needs

プランThis capability注記
無料Not on this planWorks as it always has, with standard results.
Personal Core, Personal Plus含まれるPart of Context management.
Business Core, Business Plus, Business Max含まれるPart of Context management.
Enterprise Core, Plus, Max含まれるPart of Context management.

Why it is on paid plans: your Network Agent does the heavy lifting next to your database, so the rows stay in your network and your AI client spends its attention on the thinking. That groundwork is what the paid plans add.

The Free plan works as it always has. Paid plans add the Analytic toolset. Free is a plan you keep: MCP, the REST API, federated queries and alerts, with the same results as before. Plan names and prices are on the 価格ページ.

How to ask

You do not call tools yourself. Ask in plain words and your AI client picks the tool. Three asks that bring this capability in:

For people who read tool lists: mode: shape の上 qs_run_query そして qs_run_saved_query. The answer has a shape block with the row estimate and where it came from, tokens per row, the whole-result estimate, whether it fits the budget and how many rows do, and the empty columns. See it on the Tools tab.

What comes back

Row estimate
From the database’s own catalogue when it keeps one, otherwise a count that stops early and says so. Approximate figures are reported as estimates.
Tokens per row
Measured from the first rows, so your AI client can work out how many will fit.
Fits the budget
Whether the whole result fits the size this call would use, and how many rows do.
Empty columns
Columns with no value in any sampled row, by name.
Timing
How fast the database delivered the sample and roughly how long a full scan would take, so a later profile can be sized too.

What it does not help with

  • It does not answer the question itself. It tells your AI client how big the answer is.
  • The row count of a join or a filtered query is an estimate or a capped count, not an exact total.
  • Tokens per row come from a sample. A table whose later rows are much wider will cost more.
  • A federated saved query cannot be shaped.

制限

  • A few hundred tokens for a narrow result; a wide one costs more because the column list is the answer.
  • The statement does start on your database before the agent stops it, so a shape needs the key’s run-queries scope.
  • The usual query timeout applies: 60 seconds by default, at most 300. A shape finishes quickly.

設定

No setting of its own. The rows that fit follow your Result size on the MCP page’s Context tab, card What reaches your AI assistant.

When your AI client says no

When the server refuses, your AI client receives a code and one sentence, and usually relays it. Quoted sentences are the server’s own words; the parts in italics stand for the ids a real message names.

コードWhat the server says何をするか
MCP_FEATURE_NOT_INCLUDED“Your plan does not include context management; schema by business area, shaped results and a size budget are available on Personal Core.”Your AI client carries on without the input and gives you the plain result. To keep the capability, move up to Personal Core.
MCP_AGENT_VERSION_TOO_OLD“Update the Network Agent to shape results at the source; this agent cannot answer mode=shape yet, so ask for rows (omit mode) until it is updated.”Agents update themselves. Check the agent under データ → データエージェント.
MCP_INVALID_INPUTA mode other than rows, shape, profile or quality, or a federated saved query asked to shape.Ask for rows, or shape a single-source query.
MCPエージェントが接続されていませんThe Network Agent that owns the connector is not connected right now.開く データ → データエージェント and start the agent on its computer.

例

“Is the daily inventory table safe to query?”

Your AI client learns it holds about 60 million rows and suggests a profile of the first million rows, or a monthly aggregate, instead of reading it.

“Which columns can we ignore?”

The shape names the columns that are empty in every sampled row, so the client leaves them out of the next query.

まだ行き詰まっていますか？ 開く サポート クエリ ストリームで、どのステップで何が表示されているかをお知らせください。接続をご案内いたします。

Facts source

Facts on this page: the Query Streams MCP server and Network Agent as of 19 October 2026, with the server’s own refusal sentences quoted word for word. Measured figures come from a test ERP database of 60 to 80 million rows per table. Table and column names in the examples are made up.

Setup trouble?

  • Agent shows Offline, or Test connection and Browse are greyed out? Check that the Agent is online → Is the Agent online?
更新日 10月 18, 2026

あなたの気持ちは？

  • ハッピー
  • ノーマル
  • 悲しい

搭載 ベタードック