Multiplayer Fishbowl: working several databases from one chat

Some businesses run more than one Fishbowl database: one per company after an acquisition, one per warehouse that was set up on its own, or one per brand. The systems don’t talk to each other, so anything that crosses them is done by hand. Someone enters an order in one system and the matching order in the other, and someone exports both to Excel to compare them. With ILC MCP, you can connect an AI agent to each Fishbowl in the same Claude or ChatGPT chat, and the chat works across all of them.

The setup

Each Fishbowl gets its own ILC MCP connector, and you add all of them to one chat: Site A, Site B, and so on. Each connector signs in to its own Fishbowl with your login for that system, so you only see and do what you’re allowed to in each one.

This is the same setup as a clean rebuild, where one chat is connected to an old Fishbowl and a new one. There, data moves one way, from the old system to the new. With several live systems, the agent works across all of them every day.

Pattern 1: site-to-site orders

When Site A needs stock that Site B has, the order exists twice: a purchase order in A, with B as the vendor, and a sales order in B, with A as the customer. Then B picks, packs, and ships it, and A receives it. Each step is done in a different system, often by different people, and the two orders end up with different quantities or dates.

With the steps written down once as a flow in your company’s notes, you can say:

Run the site-to-site ordering flow from SO 12345 in Site A.

The agent reads the order in Site A and then:

1. Creates a purchase order in Site A for Site B.
2. Creates the matching sales order in Site B, with the same parts and quantities.
3. Walks Site B’s team through picking, packing, and shipping the order.
4. Receives it into Site A when it arrives.

Each step shows a preview first, and someone approves it in the system it affects. Creating the orders runs on our demo system today. The pick, pack, ship, and receive steps are built and haven’t been tested end to end yet, so the full flow is early access.

Pattern 2: analysis across sites

Comparing two Fishbowl systems usually means two exports and an afternoon in Excel. When both are connected to the same chat, the agent queries each one and puts the answers side by side. This part is read-only and works today.

Some questions it can answer:

• Which parts are short in Site B and overstocked in Site A?
• What does each site pay for the same part, and from which vendor?
• Do the purchase orders Site A sent to Site B match the sales orders Site B created?

Part numbers and customer names often differ between sites. The agent can match them by description or vendor part number, show you the matches, and you confirm the ones it isn’t sure about.

The third question is a reconciliation. It finds orders that exist on one side but not the other, and quantities or prices that don’t match.

What’s there today

	Status
One chat connected to several Fishbowl systems	Works today
Read-only questions and reconciliation across sites	Works today
Creating the PO and SO for a site-to-site order	Demo system, early access
Pick, pack, ship, and receive steps	Built, not yet tested end to end

The flow lives in your company’s notes, and the agent follows those steps. Rules that must never be broken, such as which site may sell to which, are better enforced in a custom tool than written in a note.

If you run more than one Fishbowl and spend time keeping them in step, tell us which flow you’d start with.

Related:

• When your Fishbowl database does not spark joy
• What is agentic ERP, and what would it look like with Fishbowl?
• ILC MCP for Fishbowl