AI assistants

Use Claude, ChatGPT, Gemini or another AI assistant to plan, capture and review your work in LockTFI.

What it does

LockTFI can be connected to the AI assistant you already use. Once connected, you can talk to it about your work — “what should I do today?”, “turn these notes into tasks” — and it reads and updates LockTFI for you. Your board, backlog, milestones and Today change straight away, on every device you're signed in on.

You decide what each assistant can see and change. Everything it changes is listed in LockTFI, and you can undo it.

Under the hood this uses the Model Context Protocol (MCP), an open standard for connecting AI assistants to apps, so any assistant that supports MCP can connect.

Before you start

Sign in to LockTFI. Assistants work with your synced account. If you use LockTFI without signing in, your data lives only in that browser, and no assistant can reach it.

You'll need this server address when connecting:

/api/mcp

Connect your assistant

There are two ways to connect. Most assistants (Claude.ai, ChatGPT, Claude Code, Cursor, VS Code and Codex) only need the address above: they send you to a LockTFI sign-in page where you choose permissions. Tools that can't sign in, such as Gemini CLI or a script, use an access token that you create in Settings → Integrations.

{
  "mcpServers": {
    "locktfi": {
      "type": "http",
      "url": "/api/mcp",
      "headers": {
        "X-LockTFI-Project": "<project name>"
      }
    }
  }
}

{
  "mcpServers": {
    "locktfi": {
      "httpUrl": "/api/mcp",
      "headers": {
        "Authorization": "Bearer <your token>"
      }
    }
  }
}

{
  "mcpServers": {
    "locktfi": {
      "url": "/api/mcp"
    }
  }
}

{
  "servers": {
    "locktfi": {
      "type": "http",
      "url": "/api/mcp"
    }
  }
}

codex mcp add locktfi --url /api/mcp
codex mcp login locktfi

Authorization: Bearer <your token>

Keep tokens private. A token is shown only once, when you create it. Anyone who has it can do what that connection is allowed to do, so treat it like a password. If you lose it or it leaks, revoke that connection and create a new one.

Permissions

Each assistant you connect has its own permissions, which you can change at any time in Settings → Integrations. Changes apply to its next request.

Each area can be None (the assistant can't see it), Read (it can look but not change anything), or Read + write. New connections start as read-only everywhere.

Projects.Limit an assistant to chosen projects and everything else becomes invisible to it, including tasks that aren't in any project. For example, a work assistant that never sees your personal projects.

Riskier actions are off until you turn them on. Move tasks to the Trash lets it delete tasks, which stay restorable from the Trash for 30 days. Without it, an assistant can still remove tasks it created itself, so it can clean up its own mistakes. It can never delete anything permanently. Change status of Jira-linked tasks matters for projects that push status to Jira: with it on, completing a task can also move the real Jira issue.

Most records per changecaps how much one request can touch. A bigger change is refused, so a misunderstanding can't rewrite your whole backlog at once.

Things to ask

Talk to your assistant the way you would to a colleague who can see your LockTFI. Some ideas, and the permissions they need:

Some assistants also list ready-made LockTFI prompts: Plan my day, Weekly review, Triage my backlog, Break this down, Work through agent tasks, Push TODOs into LockTFI and Kick off a milestone. In Claude, look for them in the + menu; in Claude Code, type /mcp__locktfi__ to list them.

Handing tasks to an agent. Set a task's Who to Agent (or type @ai when you add it) and fill in its Agent brief: what to do, what you expect back, and the checks it must pass. Put it in the Board's Planned Agentlane to hand it over — the lane's order is the order a coding agent such as Claude Code takes them. The agent moves each to In progress while it works, then hands back a result — a summary, its deliverables and an answer to each check — and the card reads Ready for review. You complete it; the agent never does. It stops to ask when a task needs a decision, and blocked tasks are skipped. The task's AI prompt, its milestone's brief and its links are all handed over, so the more of those you fill in, the less it has to guess.

If your assistant is also connected to other apps, it can combine them. For example: “File the actionable emails from clients as LockTFI tasks”, or “Fit tomorrow's planned tasks around my calendar.”

Seeing and undoing changes

Every change an assistant makes appears under Assistant activity in Settings → Integrations. It shows which assistant made the change, what it did and when. Click Undo to put things back as they were.

If you've edited something again since the assistant changed it, Undo leaves your newer edit alone and tells you. Changes made in Jira can't be undone from LockTFI, and the activity list marks them. You can also ask the assistant to “undo that”, which reverts its most recent change.

Staying in control

Revoke a connection in Settings → Integrations to cut it off immediately. The next request it makes is refused. Its past changes stay in the activity list so you can still undo them.

Tokens you create can expire after 30 days, 90 days or a year, or never. Connections made by signing in (Claude.ai, ChatGPT, Claude Code) renew themselves while you keep using them, until you revoke them. Signing in again from the same app updates its existing connection instead of adding another.

Assistants only see what you've allowed, and only when you ask them something. LockTFI never sends your data to an assistant on its own.

Troubleshooting