BLog
/
/
What's New in PopdockAI: Resources, Rate Limits, Shared Instructions, and Four New Connectors
Content
↳
Example H2
↳
Example H3
↳
Example H4
↳
Example H5

It's a familiar afternoon for whoever looks after the company's AI setup. You open a tool's client instructions and paste in the same paragraph you pasted into four other tools last week: "Our fiscal year starts in October. Period 1 means October, not January." Someone in finance has asked the assistant what "backorder hold" means, and it guessed. And somewhere in the account, an agent is working through every open sales order one call at a time, and nobody is watching.

None of that is a disaster on its own. Together, it's the gap between an AI pilot and something you'd hand to the whole company. You need AI that knows your vocabulary, gets the same instructions wherever it works, and has a ceiling on how much it can do.

This release closes that gap with three features: MCP resources, rate limits and shared instructions. It also adds four connectors: Procore, DuckDB, Contivio and Xano. Here's what each one does and when you'd use it.

MCP Resources: Give AI the Reference Material, Not Just the Buttons

Tools let AI do things. They run a query, create a record or move a process along. What tools don't do is tell the AI what your business means by its own words. Resources do that. They live on the same MCP server as your tools and prompts, so any client that can reach one can reach all three.

There are three types:

  • File. Point a resource at a document your connectors can already reach, like a price list, a policy or a contract, and AI reads it where it is. Nothing gets downloaded and uploaded again, and no copy goes stale in a chat history.
  • Favorite. Publish a saved list view, with its filters, sort and columns, as reference material. "Aged receivables over 60 days" then means exactly what your team decided it means.
  • Glossary. A set of terms and definitions you control. Each term can point AI at the tool that answers questions about it. Set the glossary to Always on for context every conversation needs, On demand for AI to read only when it's relevant, or Off while you're still drafting.

Resources get the same governance as everything else on the server. Each resource has a version history, so you can draft a change, publish it, and roll back if it's wrong. Access is granted by user or by team, resource by resource. Every read is recorded in the resource's history. When you update the document, the favorite or the definition once, every conversation in every client picks up the change.

Our resources series covers each type in depth, including which AI clients support resources today.

Rate Limits: A Ceiling on How Fast, Not Just How Much

Your monthly task allowance already caps how much the account can do. It can't tell a busy month from a runaway agent. Ten thousand calls spread across a hundred people is healthy use. Ten thousand calls from one loop in an afternoon is not, and the allowance only notices once it's gone.

A rate limit is a rule: no more than this many calls per minute, per hour or per day. You can set one in five places, and every matching rule applies at once:

  • Default rate limits, on the Account page. These are given to every user individually, as a safety net for the whole account.
  • A user. Give a power user more room, or an automation account less.
  • A team. One allowance shared by everyone on the team, so one department can't crowd out another.
  • An MCP client. A cap on a whole rollout, like the AI assistant the support team uses.
  • An MCP tool. A cap for a tool that's slow, expensive on the source system, or writes data back.

When a limit is reached, the AI client gets a plain sentence it can pass on: "Rate limit reached. Your team is limited to 500 MCP tool calls per hour, shared between its members. Try again in 12 minutes." API callers get the same message with an HTTP 429 status and a Retry-After header. Refused calls don't spend a task, so a limit set a little low costs you a short wait, not money. Code mode scripts are counted call by call, so one script can't be used to get past a per-tool limit.

A sound place to start is a per-minute default on MCP tool calls, plus a tool limit on every tool that creates, updates or deletes records. Our rate limiting series walks through where each limit belongs and how to roll AI clients out safely.

Shared Instructions: Write the Rule Once, Use it Everywhere

Client instructions and prompt messages tell AI how to behave with a particular tool or job. They work well until the same paragraph shows up in a dozen places. When the rule changes, someone has to find every copy, and the one they miss is the one that gives a wrong answer.

A shared instruction is a block of text with a name, a description and its content, and it's kept in one place. You find them on the Shared instructions tab of the MCP page. When you write a tool's client instructions or a prompt's messages, you pick a shared instruction in place of typing the text again. Popdock puts in the current content each time the tool or prompt is served, so an edit to the shared instruction reaches every place that uses it.

The Usage table on each shared instruction lists every tool and prompt that uses it, with the MCP server it sits on. That makes it easy to check how far a change will reach before you make it.

Good candidates are the things every tool should agree on:

  • Fiscal calendar rules, like when the year starts and what a period number means.
  • Formatting conventions, like currency, date format, and how to show a customer.
  • Guardrails, like "never create a record without confirming the amount with the user first".

Four New Connectors, and What AI Can Do With Them

This release brings four new systems into Popdock: Procore for construction teams, Contivio for contact centers, Xano for teams that build their own apps, and DuckDB for history and exports kept as files. Each one below starts with the person who'll use it most and the questions they can now ask.

Procore: Run the Job from the Conversation

A project manager starts the week by finding out what's stuck. In Procore that means filtering the RFI log, then the punch list, then exporting change events to a spreadsheet to add up the exposure. With Procore tools on an MCP server, they ask:

  • "Which RFIs on the Riverside project are overdue, and who's holding each one?"
  • "What's our total exposure on pending potential change orders across all active projects?"
  • "List the open punch items on Level 3, grouped by vendor."

The AI can act on the answer as well as read it. RFIs, punch items, submittals, potential change orders, change order requests, direct costs and budget changes all have add and update actions. A superintendent can turn walk-through notes into punch items, or a project manager can draft a potential change order, without leaving the conversation.

The bigger win is across systems. Procore holds the job's commitments and costs, and your ERP holds what was actually invoiced and paid. With both connected, a construction accountant can ask, "Which Procore commitments on Riverside have vendor bills in Business Central that are over the committed amount?" Today that answer takes two exports and a VLOOKUP.

The three features in this release make that safe to hand out. A glossary resource tells the AI that a PCO, a COR and a CCO are three different things. A shared instruction says "confirm the amount before creating any change order", and every Procore action tool uses it. A tool rate limit keeps a loop from writing a hundred punch items.

Contivio: Put the Call Next to the Customer

A contact center supervisor's numbers live in Contivio, and the customer's story lives in the helpdesk or the CRM. Contivio's Call records, Call segments and Agent activities lists cover each interaction across voice, chat, email, SMS and voicemail. That includes the agent, queue, result, and talk and handle time. So the supervisor can ask:

  • "How many inbound voice calls were abandoned in the queue yesterday, by queue?"
  • "Which agents had the longest average handle time this week?"
  • "Show me every interaction from this phone number in the last 30 days."

Chat transcripts and recordings come through as files. With a File tool on the server, the AI can open a transcript and summarize what the customer actually asked for.

Things get more useful when you add a second connector. "This customer called three times this week. Do they have an open Zendesk ticket, and who owns it?" Or for a sales manager: "Which open HubSpot deals have had no outbound call in the last 14 days?"

Xano: Give AI your Own App's Database

If your product or internal app runs on Xano, its data has usually been out of reach of anyone who can't build an admin panel. The connector turns each Xano workspace into a company and each table into a list. A table you added this morning is ready to query by the afternoon. Every table also gets Add, Update and Delete record actions, so the AI can change records as well as read them.

That changes what your support and operations teams can do on their own. They can look up a user's record, correct a field or close an account from the conversation, without a ticket to engineering. Across connectors it goes further. "Which users are still marked active in our app but have a cancelled subscription in Stripe?" The AI finds them in both systems, shows you the list and, once you confirm, updates each record in Xano.

DuckDB: Keep the Old System's History Next to the New One

When a company moves off an ERP like Dynamics GP, years of history usually end up in an export nobody can query. That leaves the old server running "just in case". The DuckDB connector reads Parquet and CSV files where they already sit, in Azure Blob Storage, Amazon S3 (or an S3-compatible store) or Microsoft Fabric OneLake. Any file in the container shows up as a list with no setup. Exported snapshots can carry full list definitions with labels, relationships and favorites, so archived history reads the way it did in the original system.

With both connectors on your MCP servers, the AI can bridge the move. "Compare this customer's sales for the three years before we went live on Business Central with the year since." "What did we pay this vendor in 2019 for the same part we're buying now?"

‍

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.

Try PopdockAI Free

See Documentation