Record token usage in local agent transcripts

Ideas Feature Requests
1

Feature request for product/service

Cursor IDE

Describe the request

Cursor saves every agent chat on disk at:

~/.cursor/projects/<workspace>/agent-transcripts/<id>/<id>.jsonl

These files have the messages and tool calls, but no token usage. The tokenCount field on each message in state.vscdb is also always 0, and usageData is empty on every chat.

Please write the usage into the local transcript on each assistant message, subagent transcripts included, the same way Claude Code and Codex do in their session files:

"usage": {
  "model": "...",
  "input_tokens": 0,
  "output_tokens": 0,
  "cache_read_input_tokens": 0,
  "cache_creation_input_tokens": 0
}

Filling in the existing tokenCount fields in state.vscdb would work too. Two more things would help: tool results (success or failure, exit codes for shell commands) and timestamps on subagent transcripts.

This needs to be in the local files themselves. An API, export or dashboard doesn’t solve it.

Why

Local tools that read agent history can show cost and token use per chat, per project and per model, all without anything leaving the machine. They can do that for Claude Code and Codex today. Cursor chats show up with no cost at all.

I maintain Reflex GitHub - sumervaid/reflex-tui · GitHub , a local CLI that reads Claude Code, Codex and Cursor history to find repeated workflows. For Cursor it can already list sessions, models and launched agents, but not tokens or cost.

Thanks!

Operating System (if it applies)

Windows 10/11
MacOS
Linux