You decide what reaches your AI client.
The Context tab on the MCP page sets how every result is shaped before it leaves your network: long text, binary values, empty columns, decimals, schema for large databases, answer size, Nova, and the defaults for the deeper analyses.
How the Context tab works
Your AI client reads every answer it is handed, and its reading room is limited. The Context tab decides what that room is spent on. Your Network Agent applies the settings next to your data, so trimmed text and left-out columns never cross the wire.
Apri Streaming → In diretta → MCP, tab Context. It has two cards: What reaches your AI assistant (the organization’s result settings) and Deeper analysis (the defaults for the analysis tools). Every row has a short tooltip and a Scopri di più link that opens its own help article in place.
A setting can come from three places. For every call, the most specific one wins:
| Where it is set | Who sets it | Wins over |
|---|---|---|
| On the call | Your AI client, for one request, such as a size budget or every text cell whole. | The key and the organization. |
| On the key or connected app | An admin, with Result shaping on that key or app. | The organization. |
| For the organization | An admin, on the Context tab. | The built-in default. |
Strict keys are the exception. A key set to Strict applies the organization’s settings and a request cannot loosen them. Use it for keys handed to shared or automated clients.
- Who can change it. People with the settings permission edit both cards. Everyone else sees a one-line summary of what is on.
- When it takes effect. Changes reach connected AI clients within five minutes.
- Try before you save. IL What your AI client sees panel runs one of your saved queries through your saved settings and shows the result an AI client would receive, with what was trimmed highlighted.
- Your data never changes. Every setting shapes the answer only. Excel, Google Sheets, the REST API and Data Viewer always get the full result.
Every setting at a glance
| Collocamento | Group | Predefinito | Piano |
|---|---|---|---|
| Text cell limit | Risultati | 200 characters | Personal Core and up |
| Binary values | Risultati | Show size only | Personal Core and up |
| Drop empty columns | Risultati | SU | Personal Core and up |
| Number precision | Risultati | Spento | Personal Core and up |
| Schema view for large databases | Schema | Table list by business area | Personal Core and up |
| Large database threshold | Schema | 40 tables | Personal Core and up |
| Result size | Result size | Auto | Personal Core and up |
| Apply to Nova | Nova | Spento | Personal Core and up |
| Analysis tools | Deeper analysis | All four ticked | Business Core and up |
| Anomaly sensitivity | Deeper analysis | 0.50 | Business Core and up |
| Most groups | Deeper analysis | 8 groups | Business Core and up |
| Comparison confidence | Deeper analysis | 95% confidence | Business Core and up |
The plan it needs
| Piano | Context tab | Deeper analysis defaults | Per-key shaping |
|---|---|---|---|
| Gratuito | Not on this plan | Not on this plan | Not on this plan |
| Personal Core, Personal Plus | Incluso | Not on this plan | Included, without analysis tools |
| Business Core, Business Plus, Business Max | Incluso | Incluso | Incluso |
| Enterprise Core, Plus, Max | Incluso | Incluso | Incluso |
Gratis è un piano che mantieni. On Free the controls stay visible but locked, and results arrive the way they always have. Paid plans add the Analytic toolset: Context management and Result analysis from Personal Core, Deeper analysis on top from Business Core. Plan names and prices are on the pagina dei prezzi.
Risultati
Four settings that decide how each cell and column of an answer arrives. See also Column and cell shaping.
Text cell limit
The most characters of any text cell your AI client receives. Longer cells are cut to that length, with a marker showing how many characters were cut. The full value stays in your database.
- How it changes the answer
- Short text arrives as is. Long notes, descriptions and comments arrive trimmed, so more rows fit in one answer. Every answer that cut a cell says so and says how to get the full value, for example by asking for that one row.
- Scenario
- Ticket notes of about 2,000 characters each arrive as their first 200 characters plus the marker, so a question about last week’s tickets covers the whole week instead of a dozen rows.
- Predefinito
- 200 characters.
- Trade-off
- A lower limit fits more rows per answer but shows less of each cell. A higher limit shows more text but fewer rows fit. Raise it for contract clauses or product descriptions; lower it when you mostly count, group or scan rows.
- Choices
- 50 to 2,000 characters.
- On a single call
- Your AI client can set its own limit for one call, or ask for every cell whole for one call.
Binary values
What your AI client receives when a result contains images, files or other binary data. The bytes themselves are never sent, whichever option you choose.
- How it changes the answer
- Show size only sends a short placeholder with the size, such as
[blob 1.2 MB]. Leave the column out drops binary columns from the answer. - Scenario
- A product table stores a photo for every item. With Show size only, “which products have no photo?” works, because empty cells and size placeholders are easy to tell apart.
- Predefinito
- Show size only.
- Trade-off
- The placeholder takes a little room but tells your AI client a file exists and how large it is. Leaving the column out saves that room and hides the fact that the data is there.
- Choices
- Show size only, or Leave the column out.
- On a single call
- Not set per call.
Drop empty columns
Leaves out any column that is empty in every row of an answer. The dropped columns are named in a short note, so your AI client knows they exist.
- How it changes the answer
- Only columns that hold data take up room. More rows fit, and your AI client is not distracted by blank fields.
- Scenario
- An item table has 90 columns and 30 are never filled in at your company. Your AI client receives the 60 that hold data plus one note naming the other 30.
- Predefinito
- On.
- Trade-off
- The check looks only at the rows in that answer, so a column empty there may hold data elsewhere; the note says so. Turn it off when every answer must keep the same columns, for example a fixed report layout.
- Choices
- On or off.
- On a single call
- Not set per call. Your AI client can still name the exact columns it wants.
Number precision
Rounds decimal numbers to a set number of significant digits in the answers your AI client receives. Spento sends every number as stored. The values in your database never change.
- How it changes the answer
- Long decimals arrive shorter and easier to read, and each row takes a little less room. Totals and averages your query computes in the database still use the stored values; only the answer is rounded.
- Scenario
- A purchasing table stores unit costs such as 3.418275. At 4 significant digits your AI client reads 3.418, and a long price list arrives cleaner and shorter.
- Predefinito
- Off.
- Trade-off
- Rounding hides small differences. Keep it Off for exchange rates, measurements or any value where the last digit matters.
- Choices
- Off, or 4 to 10 significant digits.
- On a single call
- Not set per call.
Schema
Two settings that decide what your AI client sees first when it explores a database.
Schema view for large databases
What your AI client gets first when it asks for the structure of a large database. A full list of every table and column can fill a conversation before a single question is answered, so large databases start with a summary.
- How it changes the answer
- Table list by business area groups the tables by the part of the business they serve. Compact column list gives one short line per table. Dettagli completi sends every table with every column. In the summary views your AI client then asks for the tables it needs.
- Scenario
- In a database with 2,000 tables, your AI client sees groups such as Sales, Purchasing and Inventory, picks the few tables that answer the question, and asks for their columns.
- Predefinito
- Table list by business area.
- Trade-off
- Summary views keep conversations light but take one extra step to reach columns. Full detail needs no extra step and uses far more room. Choose Compact column list when your table names already say what they hold.
- Choices
- Applies only to databases above the Large database threshold.
- On a single call
- Your AI client can ask for a deeper level of any area or table as it goes. See Schema that fits.
Large database threshold
The number of tables above which a database counts as large and starts with the view chosen above.
- How it changes the answer
- A database at or below the threshold arrives as the full schema in one answer. A database above it arrives as a summary first.
- Scenario
- An accounting database has 60 tables. At the default it starts with the summary. If your AI client handles 60 tables comfortably, raise the threshold to 100 and it gets everything in one step.
- Predefinito
- 40 tables.
- Trade-off
- A higher threshold gives the whole picture at once but uses more room. A lower one keeps conversations light at the cost of an extra step. Lower it when your tables have many columns each.
- Choices
- 5 to 5,000 tables.
- On a single call
- Not set per call.
Result size
Result size
The size budget, in tokens, for each answer your AI client receives. Rows are added until the budget is reached, and a row is never cut in half.
- How it changes the answer
- Auto sizes the budget for the AI client that is asking, so a client that reads more receives more rows. Fixed keeps every answer under about the number of tokens you enter. When rows are left out, the answer says how many and how to get the rest.
- Scenario
- A question matches 50,000 order lines. The answer carries the rows that fit and says how many were left out; your AI client holds the result and pages or groups it, with no second database query.
- Predefinito
- Auto.
- Trade-off
- A larger budget shows more rows at once but fills the conversation sooner. A smaller one keeps long conversations going but needs more paging. Choose Fixed for the same answer size in every AI client.
- Choices
- Auto, or Fixed from 2,000 tokens up to your plan’s ceiling. See Client-aware defaults for what Auto picks.
- On a single call
- Your AI client can ask for a budget on a single call. See Token budget.
Nova
Apply to Nova
Whether the settings on this tab also apply to Nova, the AI analyst built into the Portal. Nova is included on Business Core and up.
- How it changes the answer
- Off: Nova keeps every value in full and the tab affects only your connected AI clients. On: the queries Nova runs follow the same text cell limit, number precision, empty-column handling and other result settings.
- Scenario
- You lower the Text cell limit to 100 characters so AI clients can scan long ticket lists. Nova users read full ticket notes in the Portal, so you leave Apply to Nova off.
- Predefinito
- Off.
- Trade-off
- On gives one rule for every AI conversation in your organization but may limit what Nova reads. Off lets Nova show results in full on screen, so answers in the Portal and in AI clients can differ slightly in detail.
- Choices
- On or off.
- On a single call
- Not set per call.
Deeper analysis defaults
The second card on the Context tab sets the defaults the analysis tools start from when your AI client runs them over a held result. A request can adjust them within these limits. These need Business Core and up.
Analysis tools
Which of the four deeper analyses your AI client may run over a held result: Anomalies, Segments, Compare e Diff. Your Network Agent computes them inside your network, without running the database query again.
- How it changes the answer
- A ticked tool can be called and returns its findings with a one-sentence summary. A tool switched off here is refused for every key.
- Scenario
- Your finance team only needs Compare and Diff. Ticking just those keeps month-end conversations on what changed and whether a difference is real.
- Predefinito
- All four ticked.
- Trade-off
- Fewer tools keep conversations focused and predictable. More tools let your AI client answer more kinds of questions, such as what is unusual or what drives a change.
- Choices
- Any combination of the four. Each key can be narrowed further, never widened.
- On a single call
- Not set per call.
Anomaly sensitivity
How readily the Anomalies tool flags a value as unusual.
- How it changes the answer
- Low: only values far from the rest are flagged. High: smaller departures are flagged too, and the list grows.
- Scenario
- A year of daily sales with one holiday spike: at 0.20 only the spike is flagged, at 0.90 ordinary busy weekends are flagged as well.
- Predefinito
- 0.50.
- Trade-off
- Lower gives shorter, more confident lists but may miss subtle problems. Higher catches more, with more false alarms for your AI client to explain. Raise it for early warning on error rates or stock levels.
- Choices
- 0.10 to 0.90.
- On a single call
- A request can adjust it within the range.
Most groups
The most groups the Segments tool may split a held result into when your AI client asks what drives a measure.
- How it changes the answer
- The answer lists up to this many segments, most important first, with a one-sentence summary.
- Scenario
- Monthly revenue fell. At the default your AI client receives the eight segments that matter most, such as wholesale in the north and online in the south, and explains the drop in a few lines.
- Predefinito
- 8 groups.
- Trade-off
- Fewer groups keep answers short and easy to act on. More give a fuller picture but use more room. Raise it when you have many regions, products or channels.
- Choices
- 2 to 12 groups.
- On a single call
- A request can adjust it within the range.
Comparison confidence
How sure the Compare tool must be before it calls a difference between two groups real.
- How it changes the answer
- The answer says whether the gap between the two groups is real or could be chance, at the confidence you set, with a one-sentence summary.
- Scenario
- Two regions average 812 and 776 per order. At 95% the tool says whether that gap is likely to hold; at 99.9% it needs stronger evidence first.
- Predefinito
- 95% confidence.
- Trade-off
- Higher confidence means fewer false alarms but may miss small, real differences. Use 99% or 99.9% when a wrong call is costly, such as pricing or staffing; 90% or 80% for early, exploratory checks.
- Choices
- 99.9%, 99%, 95%, 90% or 80%.
- On a single call
- A request can adjust it within the range.
Per-key shaping
Every MCP key and every connected app can have its own result rules. On Le mie chiavi, IL Result shaping group sits in the create and edit dialog of each key. On App connesse, open the row menu and choose Result shaping.
| Control | Choices | Cosa fa |
|---|---|---|
| Shaping | Inherit, Off, Strict | Inherit follows the organization’s Context settings. Off sends results to this client as they come from the database. Strict applies the organization’s settings and a request cannot loosen them. |
| Held results | Plan default, On, Off | Whether this client can hold a result on your Network Agent and ask follow-up questions of it. |
| Result size | A number of tokens, or blank | A size budget for this client only. Leave it blank to use the organization or client default. |
| Analysis tools | Anomalies, Segments, Compare, Diff | Limits this client to the tools picked. None picked means every tool your organization allows. Business Core and up. |
A scheduled job that must not drift
Set its key to Strict with a fixed Result size. Every run gets answers of the same shape, whatever the request asks for.
A coding client that needs raw values
Set the key’s Shaping a Spento. That client gets results as they come from the database, while chat clients keep the organization’s trimming.
Related controls outside the Context tab
- Local analysis, on the Data Agents page, per agent: turn held results on or off, set their budget and hold time, and Clear held results in one click.
- AI results, a tab on each connector: when the relationship and statistics scans run after a schema capture.
- Utensili tab on the MCP page: every tool your AI client can call, what it returns and what is on your plan. Open the Tools tab.
- Attività tab: every call, by key and by person.
- Read-only is enforced on your Network Agent whatever these settings say. Only SELECT, WITH and EXPLAIN run.
Related
Facts source
Facts on this page: the Query Streams Portal MCP page and Network Agent as of 19 October 2026, with every label as it appears on the Context tab. Table names and figures in the scenarios 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?
English 
Spanish 
German 
French 
Portuguese 
Japanese 
Korean 
Chinese 
Italian 
Dutch 
Indonesian 
Polish