Use it from your AI assistant

Use Truly Shipped from the AI assistant you already use: ask about your team's delivery, look up a pull request, or invite a member, in your own words. It works on every plan, and the assistant sees exactly what you see in the app.

What you need

• A Truly Shipped account. The assistant signs in as you, so it gets your role: see what it can do.
• An AI client that supports remote MCP servers: Claude, Claude Code, ChatGPT or Cursor, set up as below.
• The server address, the same for every workspace: https://www.trulyshipped.com/mcp

The first time you use it, your client opens a Truly Shipped page in your browser. It shows the client's name and what it's asking to do. Check them, then press Allow. Afterwards it's listed under AI clients in your account menu, where you can disconnect it. If you want an assistant that can only read, give it a read-only personal token instead.

Claude (web and desktop)

1. In Claude, open Settings → Connectors (newer versions call it Customize → Connectors).
2. Choose Add custom connector. Name it Truly Shipped, paste the server address, and press Add. Leave the advanced OAuth fields empty.
3. Press Connect and allow it on the Truly Shipped page that opens.
4. In a chat, turn on Truly Shipped from the tools menu, then ask something like “How did the Platform team's cycle time change this month?”

On Claude Team and Enterprise, an owner adds the connector once for the organization under the organization's connector settings. Each member then presses Connect and signs in as themselves. Connectors you add on the web also appear in Claude Desktop.

Claude Code

Add the server once, from a terminal:

claude mcp add --transport http truly-shipped https://www.trulyshipped.com/mcp

Then start Claude Code, type /mcp, choose truly-shipped and Authenticate. Your browser opens the Truly Shipped page; allow it and go back to the terminal. Add --scope user to the command to have it in every project, not just the current one.

ChatGPT

1. Custom connectors need ChatGPT's developer mode. Turn it on under Settings → Apps & Connectors → Advanced settings. On Business, Enterprise and Edu workspaces, an admin may have to allow it first.
2. Under Settings → Apps & Connectors, choose Create. Name it Truly Shipped, paste the server address, and choose OAuth for authentication.
3. Sign in and allow it on the Truly Shipped page that opens.
4. In a new chat, pick Truly Shipped from the tools menu.

ChatGPT's form has no field for a token, so ChatGPT always signs in through the browser. Which plans include developer mode changes from time to time; OpenAI's help center has the current list.

Cursor

Add this to ~/.cursor/mcp.json (or the project's .cursor/mcp.json):

{
  "mcpServers": {
    "truly-shipped": { "url": "https://www.trulyshipped.com/mcp" }
  }
}

Open Cursor Settings → MCP. Truly Shipped shows a Needs login button; press it and allow it in the browser. Chats started before you added it won't see it, so open a new one.

Personal tokens, for scripts and other clients

Anything that can send an Authorization header can use a personal token instead of signing in. Make one under AI clients in your account menu: it starts with tsp_, is shown once, and can be read-only or read-and-change. For Cursor:

{
  "mcpServers": {
    "truly-shipped": {
      "url": "https://www.trulyshipped.com/mcp",
      "headers": { "Authorization": "Bearer ${env:TRULY_SHIPPED_TOKEN}" }
    }
  }
}

Keep tokens in an environment variable or a password manager, never in a file you commit. Revoke one on the same page the moment it might have leaked. Deactivating a member revokes everything they connected.

What it can do

The assistant has your role's view, and nothing more. See Roles and visibility.

Free's limits apply, as they do in the app. As in the app, the numbers are context for a conversation, never a score, and the assistant is told not to rank people.

Changes: preview, then confirm

The assistant can make changes only if its sign-in included them (the Truly Shipped page lists it) or its token is read-and-change, and only the changes your role allows. Easy-to-undo changes happen straight away, like inviting a member, renaming a team or turning your digest email off. Hard-to-undo changes show a preview first and wait for your yes:

• changing a member's role, or deactivating them;
• merging people, or moving or splitting an identity;
• moving a team, which changes which managers see it;
• disconnecting a connection;
• changing your plan, with the exact amount charged now, if any;
• revoking a token.

A preview is good for 10 minutes and works once. If something changes in the meantime, you get a fresh preview instead. Admins can see everything an assistant changed, previewed or was refused under AI activity in the workspace menu.

What never goes through it

• Secrets. Jira tokens, Anthropic keys and the GitHub app install are entered on the Connections page. The assistant gives you the link.
• Card details. Paying and managing your card happen on Stripe's own pages. The assistant gives you the link.

Where your data goes

Whatever the assistant reads from Truly Shipped, it sends to its own AI provider (Anthropic for Claude, OpenAI for ChatGPT, or the model Cursor is set to), under your agreement with that provider. That processing is yours and your provider's, not ours. If your company restricts which AI tools may see engineering data, check that before connecting.

If it doesn't connect

• “Can't connect that client”: the client sent something our sign-in can't accept. Copy the message from the page and write to chris@trulyshipped.com.
• The tools don't show up: start a new chat, or reconnect (in Claude Code, /mcp → reconnect). Clients fetch the tool list when they connect.
• No change tools: the client signed in, or its token was made, read-only; or your role doesn't include those changes. Disconnect it under AI clients and connect again, or make a read-and-change token.
• “Too many requests”: each connection can make 120 requests a minute. Wait a moment.