Client-Aware Defaults: Result Sizes for Every MCP Client

Analyse your data with your AI client · Client-aware defaults

Results sized for the AI client you are using

Chat apps read results inline; coding tools can take more. The default result size follows the client, so the first try works without anyone touching a setting.

Personal Core and up Runs on your Network Agent Alleen-lezen, altijd

What your Network Agent does for you

The first query works in whichever AI client you use. When the client sets no size and your organization has not fixed one, the server picks a default that fits the client calling it: chat-sized for claude.ai, Claude Desktop and ChatGPT, larger for coding tools such as Claude Code, Cursor and VS Code.

Chat apps put every result straight into the conversation, so a large one crowds out the conversation itself. Coding tools can keep large results aside and read them in parts. One default cannot suit both, so the size follows the client:

The client is recognised from how it identifies itself when it calls. Each result says which default applied.

The plan it needs

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.

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:

Claude ChatGPT Cursor Claude Code+10 moreAlleen-lezen

Show me this year’s orders by customer.

qs_run_query (default for this client)

Use a bigger budget this time, 50,000 tokens.

qs_run_query · max_tokens=50000

Which size limit did that result use?

(reads budget_source)

Groundwork done on your Network Agent, inside your network ingelogd op het tabblad MCP-activiteit

For people who read tool lists: nothing to set. The result’s diagnostics carry budget_tokens en budget_source (client_default when this capability chose the size). See it on the Tools tab.

What comes back

Claude ChatGPT Cursor Claude Code+10 moreAlleen-lezen

List every customer with their order count this year.

qs_run_query

You have 4,210 customers with orders this year. This chat fits the first several hundred; the busiest:

klant	bestellingen
C-1042	318
C-2281	296
C-0057	271

The rest are one link away in Query Builder. In a coding tool the same question returns several times as many rows.

Groundwork done on your Network Agent, inside your network budget_source: client_default

budget_tokens

The size limit that applied to this result.

budget_source

client_default when the client decided, org when your Result size setting decided, input when your AI client set it on the call.

What it does not help with

• It does not change your client’s own limits. If a client holds less than its default, set a fixed Result size for your organization or for that key.
• An unrecognised client gets the general default, which may be smaller than it could take.
• It does not apply to Free, which returns results exactly as before.

Grenzen

• A size your AI client sets on a call always wins, then a fixed Result size, then the client default.
• The smallest size is 2,000 tokens.
• The defaults above are today’s values and may be tuned as clients change.

Instellingen

Open the MCP page’s Context tab, card What reaches your AI assistant, group Result size: Auto (the default) uses the per-client sizes above. Fixed replaces them with one number of tokens for every client. Each key can differ: when you create or edit a key on the MCP page, its Result shaping section sets Inherit (follow the organization), Uit of Strict (a request cannot loosen the organization’s settings), plus the key’s own Result size.

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.

Voorbeelden

Related

• Token budget
• Column and cell shaping
• Shape first
• Analyse your data with your AI client: every capability
• Connect Query Streams to your AI client via MCP: setup, permissions and troubleshooting

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. Table and column names in the examples are made up.