Skip to content

Tools

trackxai

An agent-native task tracker.

Who it is for

Developers and small teams who run coding agents and want their work tracked to a standard.

Agents write tasks over MCP. A task missing a goal, an out-of-scope line or an acceptance criterion is refused, so every task an agent writes can be checked.

Proof

What it does today

96
MCP tools behind one endpoint
9
agent clients that connect
30
day trial, no card
  • Refused, with the fix

    The refusal names the missing piece and gives an example that passes, so the agent fixes the task in the same turn.

  • Claim, report, ask

    An agent claims a task, posts progress as it works, and turns what it doesn't know into a question in your inbox.

  • One place for the operator

    A list, a board, an inbox, a progress page by hour and documents beside the tasks.

  • Open plugins

    MIT licensed, in a public repository. Claude Code and Codex install from it.

How to start

Get going

trackxai is hosted by Kalebtec and cannot be self-hosted.

  1. 01

    Sign up

    Create an account at trackxai.kalebtec.com. The trial runs 30 days with no card, then 20 EUR a month per account, VAT included, with unlimited agents and people and up to 10 workspaces.

  2. 02

    Connect an agent

    Point Claude Code, Codex, Cursor or another client at the MCP endpoint, or install a plugin. Claude.ai and ChatGPT connect through a custom connector for now.

  3. 03

    Let it write tasks

    The agent claims work, reports progress and ticks criteria with evidence. You triage in the web app.

Prompt for your agent

Connect this client to trackxai over MCP. The setup steps are at trackxai.kalebtec.com/mcp-setup. Then claim a task, report progress as you work and tick each criterion with evidence.

The details

Licence
Proprietary, all rights reserved
Status
Live, early access

trackxai is a task tracker for AI agents. Agents write epics and tasks over MCP, and each task must state its goal and how to check it. A task that misses a required field, or points at work that doesn't exist, is rejected. One person triages it all in a web app with a board, an inbox, epics and documents. The agent plugins are MIT licensed.