# Connect ILC MCP to Fishbowl and ChatGPT

> Connect ChatGPT to your live Fishbowl data with ILC MCP: add a custom MCP server under Plugins, sign in with your own Fishbowl login, grant read-only scopes, and ask questions in plain English. Here's every step.

ILC MCP connects ChatGPT to your live Fishbowl database so you can ask about your inventory, orders, or customers in plain English and get answers from the actual numbers. This walks through the connect, step by step, so you can self-serve.

If you're not sure what an MCP is or why you'd use one, start with [What is an MCP Query?](/blog/ai-mcp-queries-fishbowl) and [When not to use an MCP](/blog/when-not-to-use-an-mcp).

## Before you start

You'll need three things:

- **Your MCP URL.** ILC hosts the server and sends you your endpoint. In the steps below it looks like `https://yourmcp.example.com/mcp` — use the real one we send you.
- **Your own Fishbowl login.** You sign in as yourself. It's never a shared key, and your Fishbowl permissions carry over.
- **A ChatGPT account.** Any plan can read. See the plan note at the end if you want the read-write tier.

## Connect it

1. **Find the directory.** Open **Settings → Plugins → Browse directory**.
2. **Add the server.** In the plugin directory, open **Add ▾** and choose **Add custom MCP server**.
3. **Point it at ILC.** Give it a name, paste your **MCP URL**, set **Authentication** to **OAuth**, and confirm. ChatGPT shows a warning that custom servers introduce risk — that's expected for any custom connector. Connect only to servers you trust; ILC's is one you've arranged with us.
4. **Connect in one click.** Your admin pre-approved the server, so connecting takes a single click.
5. **Sign in as you.** You land on ILC's sign-in page and log in with your own Fishbowl user. Not a shared account.
6. **Grant read-only scopes.** You'll approve `sql:read` and `docs:read` — read your Fishbowl data with SQL, and read prior-session notes plus search Fishbowl's help. Nothing can write.
7. **Ask.** Open a new chat. Ask in plain English — "how many customers do we have?", "what's on hand for part FG2200?" — and the MCP writes the SQL and answers live.

Once connected, ChatGPT runs the query without prompting you for each call, so answers come back in one turn.

For OpenAI's own version of these steps, see their help article on [custom MCP connectors](https://help.openai.com/en/articles/12584461).

## What's protected

- **Read-only.** The read tier grants `sql:read` and `docs:read` only. ChatGPT can look, not change.
- **Your login, your permissions.** You authenticate as yourself through a one-time OAuth sign-in. What you can see in Fishbowl is what the AI can see.
- **Every query logged.** Each call is recorded, so there's always a record of what was asked.

## Read-only vs. read-write

ILC MCP comes in a **read-only** tier and a **read-write** tier. The steps above use read-only. Read-write grants scopes that can change Fishbowl data, so treat it with the same care as any account that can edit your system.

One ChatGPT-specific caveat for read-write: OpenAI limits write-capable connectors to **Business, Enterprise, and Edu** workspaces. On **Plus** and **Pro**, custom connectors are read/fetch-only — which is all you need for the read tier, but means the read-write tier needs a Business, Enterprise, or Edu workspace to actually write.

Prefer Claude, Gemini, or your terminal? We have guides for [Claude.ai](/blog/connect-fishbowl-mcp-claude), [Gemini](/blog/connect-fishbowl-mcp-gemini), [Claude Code](/blog/connect-fishbowl-mcp-claude-code), and [OpenCode](/blog/connect-fishbowl-mcp-opencode).

---

*From Israel Lopez Consulting (ILC). Custom Fishbowl Inventory integrations and software, and Fishbowl experts since 2015. Written by Israel Lopez, published October 28, 2026. Questions about this topic, or want to reach the people who wrote it? Visit https://israellopezconsulting.com/contact*