Use with an AI assistant · Multihub Reporting API

Ask questions about your properties in plain English and let an AI assistant run the queries. No code to write and nothing to install beyond the assistant you already use: one token, one folder, one file.

How it works

Every tool on this page can run commands on your computer. That is all it takes: you give the assistant a short instructions file that explains the API, and it fetches the data itself, reads the rows, and answers you. The instructions file is the same for every tool. Only its filename changes.

This is not an integration you install, and there is no "Multihub connector" to add. The assistant talks to the same endpoint any program would, using a token that stays on your computer.

You need three things: a Multihub API token, one of the tools below installed and signed in, and a folder on your computer to work in.

Step 1: store your token

Create a token in Multihub under Profile → API Tokens. The full value is shown once, at creation. Copy it. Then put it in an environment variable, which is a named slot on your computer that programs can read. The assistant refers to the slot by name, so the token never has to appear in a chat.

macOS

Open Terminal (press ⌘+Space, type "Terminal"). Paste this line with your token in place of the placeholder, and press Return:

echo 'export MULTIHUB_TOKEN="paste-your-token-here"' >> ~/.zshenv

Windows

Open PowerShell (press the Windows key, type "PowerShell"). Paste this line with your token in place of the placeholder, and press Enter:

[Environment]::SetEnvironmentVariable("MULTIHUB_TOKEN", "paste-your-token-here", "User")

Then quit and reopen whichever tool you are about to use, so it picks up the new variable. This is the step people most often skip.

Treat the token like a password. It carries the full read access of your account. Do not paste it into a chat window, a shared document, or a file inside your project folder. If you think it has been exposed, delete it on the API Tokens screen and create a new one; the old one stops working immediately.

Step 2: make a folder

Make an empty folder somewhere sensible, for example Documents/Multihub reports. The assistant will keep its instructions file there and, if you ask it to, save spreadsheets and charts there too. Every tool below starts by opening this folder.

Step 3: set up your tool

Each section ends with a message to paste as your first prompt. That prompt asks the assistant to download the instructions file, save it under the name the tool reads automatically, and prove the connection works by listing your datasets and property names. The assistant will ask permission before it runs commands. Approve them.

Claude Code

Open Terminal (macOS) or PowerShell (Windows) and move into your folder:
```
cd ~/Documents/"Multihub reports"
```
Start Claude Code:
```
claude
```
Paste this as your first message:

Download https://api.multihub.io/assets/docs/multihub-agent-instructions.md and save it
in this folder as CLAUDE.md. Then follow it: list the datasets I can query and the names of
my properties, and tell me what you found.

Claude Code reads CLAUDE.md automatically every time you start it in this folder.

Codex

Codex runs commands inside a sandbox that blocks the network by default, so it cannot reach the API until you allow it. Once, before you start:

Open the file ~/.codex/config.toml in a text editor (create it if it does not exist) and add these two lines:
```
[sandbox_workspace_write]
network_access = true
```
Open Terminal or PowerShell, move into your folder, and start Codex:
```
cd ~/Documents/"Multihub reports"
codex
```
Paste this as your first message:

Download https://api.multihub.io/assets/docs/multihub-agent-instructions.md and save it
in this folder as AGENTS.md. Then follow it: list the datasets I can query and the names of
my properties, and tell me what you found.

Codex reads AGENTS.md automatically. If it still reports that network access is restricted, switch its approval mode to Full access for this folder, or use the Codex extension inside VS Code, which runs commands in the editor's terminal.

Gemini CLI

Open Terminal or PowerShell, move into your folder, and start Gemini:
```
cd ~/Documents/"Multihub reports"
gemini
```
Paste this as your first message:

Download https://api.multihub.io/assets/docs/multihub-agent-instructions.md and save it
in this folder as GEMINI.md. Then follow it: list the datasets I can query and the names of
my properties, and tell me what you found.

Gemini CLI reads GEMINI.md automatically. Type /memory show to confirm it has loaded.

Cursor

In Cursor choose File → Open Folder and pick your folder.
Open the chat panel and make sure the mode is Agent, not Ask.
Paste this as your first message:

Download https://api.multihub.io/assets/docs/multihub-agent-instructions.md and save it
in this folder as AGENTS.md. Then follow it: list the datasets I can query and the names of
my properties, and tell me what you found.

Cursor reads AGENTS.md from the folder root automatically. Commands appear in the chat with a Run button; click it when asked.

VS Code

These steps are for GitHub Copilot in VS Code. If you use a different assistant inside VS Code (Claude Code, Codex or Gemini extensions), follow that tool's section instead. They read their own files regardless of the editor.

Choose File → Open Folder and pick your folder.
Open Chat and pick Agent from the mode menu at the bottom of the panel.
Paste this as your first message:

Download https://api.multihub.io/assets/docs/multihub-agent-instructions.md and save it
in this folder as .github/copilot-instructions.md, creating the .github folder. Then follow
it: list the datasets I can query and the names of my properties, and tell me what you found.

Copilot reads .github/copilot-instructions.md automatically.

Windsurf / Zed

Both editors read AGENTS.md from the folder root and treat it as always-on instructions.

Open your folder in the editor.
Open the agent panel: Cascade in Windsurf, or the Agent Panel in Zed.
Paste this as your first message:

Download https://api.multihub.io/assets/docs/multihub-agent-instructions.md and save it
in this folder as AGENTS.md. Then follow it: list the datasets I can query and the names of
my properties, and tell me what you found.

Claude Desktop

The chat side of Claude Desktop cannot run commands on its own, so there are two routes.

Code tab (recommended). Claude Desktop includes Claude Code as a tab. Click Code, open your folder, and follow the Claude Code steps, pasting the same first message. Everything else on this page applies.

Chat with the Chrome extension. If you have the Claude in Chrome extension installed, Claude can drive the playground in your browser. Open the playground in Chrome, enter your token once (it stays in that browser), then ask Claude in a chat to open the playground tab and run a query for you. Claude types the query and reads the result on screen. Slower than the Code tab, but nothing to set up beyond the extension.

Step 4: ask questions

After the first prompt the assistant knows your datasets and property names. Ask in plain language and be specific about the period. Some prompts that work well:

"What did we spend on Google Ads per property last month, and how many leads did each get?"
"Compare organic website sessions for June and July across all properties. Which ones dropped?"
"Which search keywords drove the most clicks to Riverside Commons in the last 90 days?"
"Show me review counts and average rating by property for the year to date."
"Put last quarter's paid spend, clicks and conversions by property in a spreadsheet in this folder."
"Build a chart of monthly sessions by channel for Oak Street Lofts since January."

Reading the answers

A good answer names the dataset it used and the exact date range; the instructions file asks for both. The assistant is also given the caveats in Before you build, including that data ends three days ago. If a figure does not match what you expect, ask it to show the query it ran.

Keeping it safe

The token is read-only on this API: the assistant can run queries, and every mutation is refused. It stays on your computer, in the environment variable, and the instructions tell the assistant never to print it.

Rows go to the assistant's provider. Whatever the assistant fetches becomes part of the conversation, the same as anything you paste in. Use the assistant your organisation has approved for business data.

Troubleshooting

What you see	What it usually means	Fix
"Unauthorized" or "401"	The tool cannot see the token.	Quit the tool fully and reopen it. Then ask: "Is MULTIHUB_TOKEN set? Don't print its value." If not, repeat Step 1.
"Could not resolve host" or a timeout	The tool's sandbox is blocking the network.	Codex: see the config step in its section. Others: allow the command when the tool asks.
The assistant asks you for the token	It has not read the instructions file.	Do not paste it. Check the file exists in the folder under the name your tool reads, then start a new chat.
A dataset returns zero rows	Your properties have no connection feeding it.	Not an error. The assistant can tell you which integration it needs; your Repli contact can connect it.
Numbers differ from a Multihub report	Different date window, or one of the caveats above.	Ask the assistant for its date range and dataset, then compare against Before you build.
Property ids instead of names	The assistant skipped the names query.	Ask: "Run the companies query and replace the ids with names."

On this page