What Claude and ChatGPT can see and do in Korrit

When you connect Claude or ChatGPT to Korrit, the AI can read and change some of your firm’s data on your behalf. This page sets out exactly what it can reach. To set up a connection, follow Connect Korrit to Claude or ChatGPT.

How the connection works

Korrit runs an MCP server. MCP (Model Context Protocol) is an open standard that lets an AI app call a fixed set of actions, called tools, in another product. The AI can’t browse Korrit or run its own queries against your data. It can only call the tools listed below.

When you connect, you sign in to Korrit and approve the request on a consent screen that names the AI app, the firm, and the role the connection will act with. Choosing Allow gives the AI app a key tied to you in that one firm. This is standard OAuth: the AI app never sees your Korrit password, and its key expires after an hour and renews automatically while the connection stays active.

Apps that can’t sign in this way can use a personal access token instead. Korrit shows it once, stores only a fingerprint of it, and it works until you revoke it. Claude and ChatGPT don’t need one.

It acts as you, with your permissions

Every tool call runs as the person who connected, with exactly that person’s role and permissions. Each tool checks the same permission as the matching screen in Korrit, so the connection can never do more than you can do in the app yourself.

In practice:

• A designer who can only see the projects they’re assigned to gets only those projects through the AI.
• If your role can only read your own time, the AI can’t read anyone else’s timesheet for you.
• Pay rates appear only if your role holds the compensation permission. Without it, labor cost and profit figures come back empty.
• The analytics reports need firm-wide analytics access, which by default only principals have.
• A connection works in one firm. It can’t reach data in any other Korrit firm.

If your role changes, the connection follows the new permissions on its next request. It stops working if your membership is suspended, or until the firm’s subscription is active again if it lapses.

Every tool the AI can use

This is the complete list. If an action isn’t here, the AI can’t do it through Korrit.

Your account and team (read only)

• Who am I: your name, role, firm, today’s date in the firm’s time zone, and what your role may do in each area.
• List team members and Get a team member: names, emails, roles, titles, departments, offices, weekly capacity and active projects, plus one person’s hours this week if your role can read their time.

Projects (read only)

• List projects: status, client, project manager, open tasks and logged hours.
• Get project state: each phase with its status, percent complete and budgeted hours, the fee total, and the assigned team.

Tasks (read and write)

• List a project’s tasks and List my tasks (your assigned project tasks plus your personal to-do list).
• Create a task: on a project, optionally in a phase and assigned to someone, or as a private task on your personal list.
• Edit a task: title, description, status, priority, due date, phase or assignee.
• Mark a task: done, reopened, in progress or blocked.

Time (read and write)

• Log time to a project phase, for yourself or, if your role allows it, for someone else.
• Edit a time entry and Delete a time entry. Approved or invoiced time can’t be changed, and neither can time on a day whose timesheet period is marked complete or approved.
• List time entries, Get a weekly timesheet and Summarize hours by person and project over a date range.

Analytics (read only, firm-wide access needed)

• Get the profit report per project and per person.
• Get projects per quarter: active, started and completed.
• Get weekly hours: billable and non-billable time against capacity.
• Get people analytics: timesheet on-time rates and automation activity.

Does the AI ask before it changes something?

Korrit doesn’t add its own confirmation step: if your role allows a change, it’s made. Korrit does label its tools for the AI app, marking lookups as read-only and deleting a time entry as destructive.

Whether you’re asked first is up to the AI app and your settings in it. Claude asks for your approval before it uses a connector tool unless you’ve chosen Allow always for it. In Research mode it can use them without asking, so Anthropic suggests switching off tools that make changes first. Check your settings in ChatGPT before you rely on it asking.

Changes made through the AI count as your changes and start your firm’s automations just as they would in Korrit.

What data leaves Korrit

Only what a tool returns. When the AI calls a tool, Korrit sends back the result of that one request, such as a project’s phases or a week of time entries, and the AI app reads it to answer you. The connection doesn’t share anything beyond those results.

Once a result reaches Claude or ChatGPT, it’s part of your conversation there and falls under that provider’s terms and your account settings. Read Anthropic’s privacy policy and OpenAI’s privacy policy for how they handle it.

What Korrit records

• Every connection is listed under Settings → Connectors → Manage → Your connections with its name, how it connected and roughly when it was last used.
• Approving a connection, creating a personal access token and revoking a connection are written to the firm’s activity log.
• Changes the AI makes to tasks and time are logged under your name, like changes made in Korrit. The log doesn’t mark them as coming from the AI.

Lookups through the AI, such as listing projects or reading a timesheet, aren’t written to the activity log. On the Studio and Enterprise plans, principals and anyone else with firm-wide analytics access can review the log in Analytics → Activity.

Revoke access

1. Open Settings → Connectors and choose Manage on the MCP card.
2. Under Your connections, choose Revoke next to the connection.

It stops working immediately, including any access key the AI app is still holding. You can only see and revoke your own connections. For a broader look at how Korrit protects your data, read Security and data ownership at Korrit.