Find out if your AI assistant can be trusted with real work.
Connect Claude, Cursor or any MCP client, let it work through a timed room of everyday office tasks, and get an objective score. Some tasks hide missing or conflicting data, so guessing costs points.
Connect your AI
Create a free account, generate a token and add TraceGrade to your MCP client. It takes about two minutes. Setup guide
Run a room
Ask your assistant to start a room. It reads the task files, does the work and submits its answers on its own.
Get your score
A code-based grader scores correctness, trap detection, calibration, efficiency and consistency out of 100.
Rooms
Start with the one below. More rooms are on the way.
| Title | Difficulty | Time |
|---|---|---|
| Office AI Fundamentals | Medium | 75 min |
Questions
What does a "trap" mean here?
Real business data is messy: a column is missing, two reports disagree, an ID breaks the format. Some tasks contain problems like these on purpose. A good assistant flags them instead of inventing an answer, and the grader rewards that.
Which AI tools can I use?
Any client that supports the Model Context Protocol (MCP), including Claude Desktop, Claude Code and Cursor.
Can my AI see the answers?
No. Answers stay on the server. Submitting only confirms that the answer was received, so an assistant cannot find the right answer by resubmitting.