You can connect an AI assistant you already use to your Gript account, so you can ask it things like "what is overdue on the Projects board?", "add a comment saying the client approved it", or "set up a rule that moves finished work to the Done group". You approve the connection on a Gript screen in your own browser, you choose exactly what the assistant may do, and you can lower or withdraw that at any time.
This guide
An assistant that lives outside Gript, reaching in: ChatGPT, Claude, Gemini, Grok, Claude Code and others.
The AI Assistant guide
The AI built into Gript itself: Write with AI, Notebook AI Chat and Inline AI Writing. Those run on your own provider key and are set up separately. See AI Assistant.
These are two separate things and neither needs the other. Connecting an assistant does not need an AI provider key, and it does not touch your AI budget in Gript. The assistant runs on its own account, and Gript simply answers it.
1
The address every app asks for
One address, the same for everybody.
Whichever app you are connecting, it will ask you for a server address. Gript has one:
https://my.gript.app/api/mcp
Gript shows you this address inside the app, with a Copy button next to it. Open the Control Panel, click Connected AIs in the Connections group of the sidebar, then click How to connect an AI assistant at the bottom of that page. The address shown there is always the right one for the site you are signed in to.
2
Connecting, app by app
Different menus, same ending: a Gript permission screen and an Allow access button.
ChatGPT
ChatGPT needs developer mode, which is in ChatGPT's own settings.
1
Turn on developer mode
In ChatGPT, open Settings, turn on developer mode, then go to Plugins.
2
Add a new plugin
Click +, then New Plugin.
3
Paste the Gript address
Paste https://my.gript.app/api/mcp and choose OAuth as the authentication.
4
Sign in to Gript
ChatGPT opens a Gript page in your browser. If you are not signed in to Gript already, the normal Gript sign in appears first. Google, Microsoft or an email code all work.
5
Choose what it may do, then click Allow access
The Gript screen lists the permission areas, each set to None, Read or Read + write. Read section 3 below before you decide.
6
Finish in ChatGPT
Back in ChatGPT, open the plugin and click Connect if it is not connected yet. Its tools appear once the connection finishes.
7
Use it
In a new chat, open the plus menu and select Gript as an app for the conversation (under Developer mode), then ask. Attached only as a search source it can search and nothing else. Every change it makes asks you to confirm once; that is ChatGPT's own rule.
Claude, web and desktop
The steps are the same in the Claude web app and the Claude desktop app.
1
Open Customize, then Connectors
2
Add a custom connector
Click +, then Add custom connector.
3
Paste the Gript address
Leave Advanced settings alone. Claude sets itself up automatically.
4
Click Add, then click Connect
Connect is on the new connector you just added.
5
Sign in to Gript, choose what Claude may do, then click Allow access
6
Use it
Start a new chat and ask in plain words, for example "what is overdue on the Projects board?". Grant every area you want before you start: when a tool Claude calls is not allowed, it stops without saying why.
On a Team or Enterprise plan an owner adds it once under Organization settings, then Connectors. After that each person clicks Connect for themselves, gets their own permissions, and sees only their own data.
Gemini
Currently US only. The Custom apps for Spark section, where a custom app is added, appears on the Connected apps page only when Google sees your account in the United States; outside the US the section is simply absent, and there is no other place in Gemini to enter a server address. It also needs Gemini Spark with a Google AI Pro or Ultra subscription, a personal Google account, English, age 18 or over, and Keep Activity switched on. You can only add one on a computer, though it then works on mobile too.
1
On a computer, go to gemini.google.com
2
Open Settings & help, then Connected Apps
Settings & help is at the bottom of the page.
3
Under Custom apps for Spark, click Add a custom app
4
Paste the Gript address
Leave Advanced features alone. Gript registers itself automatically.
6
Follow the instructions on screen
Then choose what it may do on the Gript screen and click Allow access.
7
Use it, inside Gemini Spark only
Custom apps do not exist in the ordinary chat. Open Spark, type @, pick Gript from the list, then ask, for example "@Gript what boards can you see?".
Grok
1
Go to grok.com/connectors
5
Sign in to Gript, choose what it may do, then click Allow access
On Business and Enterprise plans a team admin has to add it once in the xAI console first, and then everyone connects their own account.
Claude Code
Claude Code is the terminal tool, so this one is a command rather than clicks.
1
Run the add command
claude mcp add --transport http gript https://my.gript.app/api/mcp
2
It is added but not signed in yet
That is expected. Nothing has gone wrong.
3
Sign in
Run: claude mcp login gript. Or type /mcp inside a session, pick gript, and choose Authenticate.
4
Choose what it may do, then click Allow access
Your browser opens the Gript screen.
5
Use it
Type /mcp in a session to see gript listed with its tools, then ask in plain words. Claude Code picks the tools itself.
Other apps
Zed, Perplexity, Windsurf, Gemini CLI and Microsoft Copilot Studio connect the same way: give them the address, follow their own add a server step, then sign in to Gript. Gript registers itself with them automatically, so there is nothing to fill in. Gemini CLI is a single command: gemini mcp add --transport http gript https://my.gript.app/api/mcp
Cursor and VS Code with GitHub Copilot are configured by editing a small settings file and then following the sign in prompt. In Cursor that file is .cursor/mcp.json, holding {"mcpServers": {"gript": {"url": "https://my.gript.app/api/mcp"}}}. In VS Code, run MCP: Open User Configuration and add {"servers": {"gript": {"type": "http", "url": "https://my.gript.app/api/mcp"}}}.
Cline and JetBrains AI Assistant can reach the address, but their own guides do not describe the sign in step, so use an API key with those instead: create one under API keys in the Control Panel and send it as an Authorization header. The consumer Microsoft Copilot app has no way to add your own server at all.
3
What each permission covers
You are never agreeing to a vague "access to your account". Every area is separate, and every one can be None.
Each area is set to one of three levels: None, Read, or Read + write. Read lets the assistant look. Read + write lets it change things. None means it cannot even see that the area exists.
AreaWhat it gives a connected assistant
Boards and itemsThe biggest area: boards, columns, sections and tasks, plus saved views, dashboards, folders, board settings, diagrams and the calendar. Read: list and read all of it, search, and pick out what is ready to work on. Read + write: create and change it, including a whole set of columns in one call and up to 100 tasks in one call.
CommentsRead: read the discussion on a task. Read + write: post a comment as you, and react to one.
FilesRead: list the files on a task, and search files across the account. Read + write: attach a file, start an upload, or create an empty Word, Excel or PowerPoint document.
NotebooksRead: list notebooks and their pages, read a page, and let pages appear in search. Read + write: write a new page, rewrite one, and comment on a page.
AutomationsRead: list your rules, read one in full with its run history, and read the catalogue of triggers and actions. Read + write: turn a rule on or off, propose a new rule or a change for you to approve, and run a rule back over tasks that already exist. See Automations.
Sticky notesRead: see your notes. Read + write: write, change, arrange, pin and send them, and set reminders on them.
Time trackingRead: what is logged on a task, the running timer, the time report and the hourly rates. Read + write: log time, run the timer, correct or delete an entry, set a rate. Logging time needs this area; Boards on its own is not enough.
PeopleRead only: the people in the account with their ids, and who is online. This is how a name becomes the id a Person column needs.
FormsRead: the forms, their fields, their submissions and their access codes. Read + write: create, change and publish a form, and make access codes.
EmailRead: connected mailboxes, the email on a task, templates, signatures, shared mailboxes and email rules. Read + write: send email once you have agreed, save a template, write an email rule.
ImportsRead only: which other tools Gript can import from, and a look at what is in your Asana, ClickUp or Jira before you run that import yourself in Gript.
Invoices and budgetsRead only: invoices and the resource plan.
Service targetsRead only: the SLA policies and their clocks. Needs an admin account.
GoalsRead: your goals and where each one stands. Read + write: create, change and delete a goal.
The trashRead: everything still recoverable. Read + write: bring something back, or empty it for good once you have agreed.
WebhooksNothing for a chat assistant, because there are no webhook tools for one. It is there for API keys, which do use it.
The connect screen offers fifteen areas, not sixteen. Webhooks is deliberately left out of what an AI app is asked to request, because a chat assistant has no use for it. The Connected AIs page lists all sixteen, so you can still grant Webhooks there later if you ever need it. Four areas are offered as Read only, because nothing in them can write: People, Imports, Invoices and budgets, and Service targets.
Three things stay true whatever you allow. Your password is never shared, and the app never even sees it. A connected app can never do more in Gript than your own account can. And what it may do is read from your approval on every single call, so lowering a level or disconnecting bites immediately, with no old token still carrying the old powers.
4
Automations is never granted automatically
The one area you always have to tick on purpose.
An automation runs by itself, over and over, and one of the things it can do is send email on your behalf. So Automations is a permission area Gript never hands out quietly. Four more are handled the same way, because each one reaches outside the account, spends money or destroys something for good: Email, Imports, Invoices and budgets, and The trash. An API key made before these areas existed reaches every other area and never these five.
WhereWhat you have to do
Connecting a new assistantSet Automations to Read, or Read + write, on the Gript permission screen before you click Allow access. It arrives preselected at whatever the app asked for, and like every other row you can turn it down to None.
An assistant you connected earlierOpen the Control Panel, click Connected AIs, click the app's row, and set Automations there. Raising a level always routes through a separate confirmation screen. Removing the connection and adding it again works too.
An API keySet Automations when you create the key, under API keys in the Control Panel. A key made before permissions existed reaches every other area but never this one, so nothing you made in the past silently gained the ability to send email.
Read is enough to look, Read + write is needed to change. An assistant with Automations set to Read can tell you what your rules do, but it cannot switch one on and cannot propose a new one.
5
Changing what a connected app may do
One page, live, with no reconnecting.
1
Open the Control Panel
Use the Settings button at the bottom of the left sidebar, the menu under your avatar in the top right corner, or press Cmd and comma (Ctrl and comma on Windows).
2
Click Connected AIs
It is in the Connections group of the sidebar, directly under API keys.
3
Read the row
Each app shows its name, where it came from, when it connected, when it last called Gript (or Never used), and a summary of its levels: No access, Read-only or Custom access. A green shield means Gript recognises the app. Anything without one registered itself.
4
Click the row to open the picker
All six areas are listed, each with None, Read and Read + write, showing what this app has right now.
5
Click a lower level to take access away
It saves immediately and applies on the app's very next call. No re-approval, no waiting.
6
Click a higher level to give more
Gript does not apply this silently. You are taken to a confirmation screen showing exactly which area is being raised, and only Allow access there applies it.
7
Click Disconnect to cut it off completely
You are asked to confirm, then the app loses access immediately, along with every token it holds. It can ask to connect again later, which starts a fresh permission screen.
6
Two things that catch people out
Both of these look like something is broken. Neither is.
Start a new chat after changing permissions. An assistant fixes its list of available Gript tools at the moment a conversation opens. If you grant it something new while a chat is already running, that chat keeps the old list and will keep telling you it cannot do the thing. Open a brand new chat and ask again. This is the most common cause of "my assistant says Gript cannot do that".
An existing connection keeps the permissions it was given. Granting an area to one app does not grant it to another, and it does not change a connection you made last week. To add something to an app that is already connected, open its row under Connected AIs and set the area there, or remove the connection in the AI app and add it again.
The two rules, side by side. A new Gript ability reaches an assistant only when it next reads the list of tools, which is when a conversation starts, so start a new chat. A permission change reaches it on its very next message, in the chat you already have open, so nothing needs reconnecting. The row under Connected AIs shows when each app last read the list.
My assistant says Gript has no automation tools.
Two causes, and it is usually the second. First, check the app really has the permission: Control Panel, Connected AIs, click its row, and look at Automations. If it says None, set it to Read + write and confirm on the screen that follows. Second, and more often, you are in a chat that was already open. An assistant fixes its tool list when the conversation starts, so open a brand new chat and ask again.
It keeps failing, and retrying does not help.
Remove the connection in the AI app and add it again from scratch. An assistant reads Gript's connection settings once, when you first add it, and keeps using that copy. If anything about the connection changed afterwards it keeps using the stale copy and keeps failing, however many times you retry. Removing and re-adding is what makes it read the current settings. This is the first thing to try in ChatGPT, where we have seen it happen.
I connected it. Now what?
Address it the way that app expects, and start small. ChatGPT: select Gript as an app from the plus menu in a new chat. Claude: a new chat, plain words. Gemini: inside Spark only, type @ and pick Gript. Claude Code: /mcp shows it, then just ask. Try "what boards can you see?" first; a whole board comes after.
ChatGPT says it can only search Gript, and cannot create or change anything.
It has Gript attached as a search source, which exposes search and nothing else. Select Gript as an app for the conversation instead, from the plus menu under Developer mode, and every tool appears. Each tool that changes something asks you to confirm it once; that is ChatGPT's own rule for write actions, not a Gript limit.
It says it is not allowed to do something, or asks me for a permission.
The connection was not granted that area. An app that follows the protocol opens Gript's approval screen with just that one area for you to allow; otherwise open Control Panel, Connected AIs, click the app's row and raise the area it named. It takes effect on the very next message, with no new chat.
Can it rebuild a spreadsheet as a board?
Yes, and it is about five calls: the board, the columns in one call, a section, all the rows in one call (up to 100 at a time), and a read back. Attach the file and ask for the board by name. If a value is in a shape the board refuses, that one cell comes back with the reason and the rest of the row is still written, so the assistant repairs the cell rather than starting again. Sending the same batch twice cannot double the board.
Why does it keep asking me to confirm things?
Two different things. ChatGPT asks once for every action that changes something, which is its own rule. Gript itself asks before anything that cannot be taken back or that reaches outside the account: deleting a column, section or board, sending an email, publishing a form, setting an hourly rate, running a rule over existing tasks, emptying the trash. The assistant shows you exactly what would happen and waits for your yes.
It says connected, but not all permissions were granted.
That is normal and nothing is broken. The app asked for more than you allowed, which is exactly what the permission screen is for. If it later tries something you did not allow, that one action is refused and everything else keeps working.
It says there are no tools or no actions available.
The connection did not finish. Tools only arrive after you click Allow access and the app collects its access, so an empty list means it stopped before that. Open the connection in the AI app, finish it, then check that it appears under Connected AIs.
It sent me to a sign in page instead of the permission screen.
You were not signed in to Gript in that browser. Sign in and it carries you straight on to the permission screen. Nothing is lost.
The permission screen showed an amber warning about the app.
It means the app registered itself with Gript and has not been reviewed by us. Gript marks an app as recognised only when it identifies itself in a particular way and comes from an address Gript knows, which today means ChatGPT and Claude. Most other apps, Gemini, Grok, Claude Code and the code editors included, will show the warning, and on its own that is not a sign of trouble. What matters is that you started this connection yourself and you trust the app. If a permission screen turns up when you were not connecting anything, click Cancel.
Can two people share one connection?
No, and that is deliberate. Every person connects their own account and gets their own permissions, so an assistant only ever sees what that one person can see. On a Claude Team or Enterprise plan an owner adds the connector once, but each person still clicks Connect and approves for themselves.
Does connecting an assistant cost anything, or use my AI budget?
No. This is not Gript's own AI, so no provider key is involved and nothing is counted against the AI budget in the Control Panel. The assistant runs on its own account and Gript just answers its requests.
How do I stop an assistant completely?
Open the Control Panel, click Connected AIs, open the app's row and click Disconnect. Access ends immediately, along with every token the app holds. Removing the connector in the AI app as well is tidy, but not required.