Clientish Documents
Step-by-step guides for every module, from your first workspace to proposals, billing, support and automation.
Connect an AI Agent (MCP)
Give Claude, Cursor or Codex read-only access to your workspace: create a connection, choose what it may read, and add it to your AI tool.
AI Connections let an AI agent such as Claude, Cursor or Codex read your workspace. You ask questions in plain language, like “Which invoices are unpaid and past their due date?”, and the agent answers from your live data. It can only read; it can never change anything.
Overview
This guide shows how to create a connection, choose what the agent may read, add it to your AI tool, and keep it safe. Clientish uses the Model Context Protocol (MCP), the standard way AI agents connect to other apps.
Who Can Access
Only the workspace owner can create and manage AI Connections.
Prerequisites
- An AI tool that supports MCP: Claude Code, Claude Desktop, Cursor, Codex or another MCP client
- Access to the AI Connections page in your workspace sidebar
Step-by-Step Guide
Open AI Connections
In the sidebar, click AI Connections, then New connection. How to connect opens the setup guide for each AI tool.
Name it and choose what it may read
Give the connection a name you will recognise, such as the device or tool it is for. Then tick what the agent may read: Clients, Orders, Invoices, Tickets, Services, Leads, Team and Reports. Everything is read-only.
Set limits (optional)
- Expires: the connection stops working at the end of that day. Leave empty for no expiry.
- Requests per minute: between 1 and 1,000 (default 120).
- Allowed addresses: IP addresses or ranges, one per line. Leave empty for a laptop, because home and café addresses change often.
Click Create connection.
Copy the connection URL
The URL is shown only once. Click Copy and keep it somewhere safe. If you lose it, issue a new URL (step 6).
Under Set up your agent, pick your tool (Claude Code, Claude Desktop, Cursor, Codex or Other) and copy the ready-made command. Then click I have copied it.
Add it to your AI tool and ask a question
Run the command or paste the URL into your AI tool, following the steps for that tool under How to connect. Then ask something like “List my open orders in Clientish.”
Manage your connections
Every connection is listed with what it can read, its limits, when it was last used and how many calls it made. Click the eye icon on a row to open it: you see its calls, refused calls, the questions it can answer and what it is not allowed to read.
At the bottom: Edit changes what it may read and its limits, Issue a new URL replaces the URL (the old one stops working), and Revoke switches the connection off.
Check the activity
Activity shows every call: the time, the connection, the tool the agent used and whether it was answered or refused. Only argument names are logged, never their values, and entries are kept for 90 days.
What the Agent Can and Cannot Do
- It can read only the areas you ticked for that connection.
- It can never create, change or delete anything. Ask it to change an order or send an invoice and it will say it cannot.
- Calls from an address that isn’t allowed, over the rate limit or outside its scopes are refused and counted under Refused, last 7 days.
Tips & Best Practices
Create one connection per device or tool, and tick only what it needs. If a laptop is lost, revoke just that connection.
Check figures before you share them. The data is live, but an AI can still add up or summarise wrongly.
Common Issues
The agent says it can’t find something: the connection may not be allowed to read that area. Edit the connection and tick it, or ask about an area it can read.
Lost the URL: it can’t be shown again. Issue a new URL from the connection’s row and update your AI tool.