
Somewhere in your business, there was a conversation about what "open orders" means. Does a partially shipped order count? What about one on credit hold? Does it include the intercompany transfers or not?
That conversation happened, someone made a call, and the result is a saved view sitting in PopdockAI with the filters set the way your team decided. Every time someone opens it, they get the answer the business agreed on.
A favorite resource publishes that view to your AI clients. Instead of the assistant assembling a query and hoping it matches what people meant, it reads the view your team already validated.
This article covers what makes that different from letting AI query the data itself, how to set one up, and how to choose between a favorite resource and a Get Data tool.
The Problem with Letting AI Decide What to Include
A Get Data tool gives an AI parameters and lets it choose the values. That is genuinely powerful, and for open ended questions it is exactly right. It also has two failure modes that are easy to miss.
The filters are plausible but wrong. Ask for open orders and a model will do something reasonable. It will find a status field, pick the values that sound like they mean open, and return a result. The number will look fine. Nobody will notice that it included the credit holds your finance team deliberately excludes, because the only way to notice is to already know the answer.
Two people get two different answers. This is the more corrosive one. The same question, asked in slightly different words by two people, produces two different queries and two different totals. Both are defensible, and neither matches the number in the weekly report. Trust in the whole setup erodes from there, and it erodes quietly, because the disagreements surface as confusion rather than as errors.
Neither problem is a shortcoming of the model. No schema can tell it that your business excludes credit holds. That knowledge lives in the decision your team made, not in the data.
The View is the Definition
A favorite already carries three things, and all three travel with it when you publish it as a resource.
- Its filters. The conditions that define membership, including the ones that only make sense if you know the business.
- Its sort order. What "most important first" means for this particular view.
- Its columns. Just as importantly, what to leave out. A view with eight relevant columns produces a better answer than a table with sixty, and the choice of eight is institutional knowledge.
Publishing that as a resource means there is nothing for the AI to fill in. There are no parameters to get wrong, because the decisions were made by the people who own them.
Setting One Up
Open the MCP server, go to the Resources tab, and choose Add resource, then Favorite. Pick the connector, pick the list, then pick the favorite. If the view already exists, this takes under a minute. There are no fields to map and no queries to write.
One thing to know up front: the connector, list, and favorite pairing is fixed once the resource is created. You can rename the resource, rewrite its description, and change its status, but to publish a different view you add a new resource. In practice this is rarely a constraint, since the whole point of a favorite resource is that it represents one specific, stable view.
Put the Definition in the Description
The description is where you write down what the view actually means, and it is the most valuable field on the page. The AI reads it to decide whether this resource answers the question in front of it.
A description that adds nothing:
- Open orders - List of open orders
A description that carries the decision:
- Open orders - Sales orders accepted but not yet fully shipped, across all regions. Excludes orders on credit hold and intercompany transfers, matching the definition used in the weekly operations report. Sorted by requested ship date. Use this for any question about outstanding order volume or upcoming fulfillment.
The second version does something the first cannot: it tells the AI what is not in the data. That is what stops an assistant from confidently answering a question the view was never meant to answer, and it is the sentence most people skip.
Favorite Resource or Get Data Tool?
Both put list data in front of AI, so the overlap is real. The distinction comes down to whether the question or the answer is fixed.
A Get Data tool is for questions you cannot predict. "Show me customers in the northwest with more than fifty thousand in orders this quarter" needs parameters, because every part of it varies. The AI supplies the values, validation checks them, and the tool runs. This is the right choice for exploration and ad hoc analysis.
A favorite resource is for an answer that should not vary. "Our open orders" has one correct interpretation in your business, and it is not up for negotiation on a per-question basis. There is nothing to parameterize because nothing should change.
The practical test: if two people asking the same question should get identical results, publish a favorite. If the results should depend on what they asked for, build a Get Data tool.
They work well in combination. A favorite resource establishes the canonical view of open orders, and a Get Data tool lets someone drill into a specific customer's orders when the canonical view raises a question.
Predictable by Design
There is a quieter benefit worth naming. An open-ended query can return anything, which means it can return an enormous amount. A favorite has a defined scope, so what comes back stays in a predictable range.
That matters in two directions. Your source systems are never asked an unbounded question, and your token costs stay stable, because the size of a resource read does not depend on how a model chose to phrase a filter.
The Maintenance Advantage, and its One Catch
Because the resource points at the favorite rather than copying it, editing the view updates every conversation that reads it. When "priority accounts" gains a fourth criterion, you change it in one place and nobody has to be told twice. This is the strongest argument for favorites over any approach that restates business logic somewhere else.
The catch is the same mechanism seen from the other side. Resource versions cover the resource itself, its name, description, and status. They do not snapshot the favorite's own definition. So if someone edits the underlying view, the meaning of the resource changes immediately, without a version being published and without appearing in the resource's version history.
That is usually what you want, since it is the whole point of pointing at a live view. But it means the favorites behind published resources deserve to be treated as governed assets. Worth knowing which of your views are published, and worth saying so to the people who maintain them.
Which Favorites the Publish First
Look for views that are already load bearing:
- The view behind a recurring report. If a number gets read out in a weekly meeting, the view that produces it is the canonical one, and publishing it means AI answers match the meeting.
- The view a team lives in. Whatever your support team leaves open all day is, by definition, their working definition of the queue.
- Anything with a contested definition. Aged receivables, active customers, at-risk renewals. Where the definition took an argument to settle, publishing it stops the argument being reopened by an AI every time someone asks.
- Views with careful exclusions. If a view exists mainly to leave certain records out, that exclusion is exactly the knowledge a model cannot infer.
The Short Version
Letting AI assemble its own query is right for questions you cannot predict, and wrong for questions that already have an agreed answer. A favorite resource publishes the view your team validated, with its filters, sorting, and columns intact and no parameters for a model to get wrong. Setup takes a minute because the view already exists, the description is where you record what the view includes and deliberately excludes, and edits to the view flow through to every conversation immediately. Use a Get Data tool when the results should depend on what was asked. Use a favorite resource when everyone should get the same number.
From AI Potential to AI Performance in Minutes
One layer.
All your apps.
Any AI tool.
Join forward-thinking companies that are transforming their operations with intelligent, secure AI automation.

.webp)
