Nice scope on this — wrapping the full Enterprise Admin/Analytics API as MCP tools is a much better interface than the dashboard for "who spent what" questions.
Question: does the Cursor API expose per-user/per-model spend at fine enough granularity — and freshness — that you could drive enforcement from it, not just reporting? Context: we build AgentShield, a pre-call spend firewall for agents (stdlib Python, per-transaction rules — velocity checks, daily caps, per-call ceilings). The gap we keep seeing is that analytics tools tell you on Tuesday that Monday was expensive, while the expensive failure (a stuck retry loop, $50–$200 in ~10 minutes) needs data fresh enough to trip a rule mid-run.
If cursor-usage can expose near-real-time per-user burn, a rule engine could subscribe to it — genuinely curious what the API's data latency looks like. https://agentshield.fly.dev
Nice scope on this — wrapping the full Enterprise Admin/Analytics API as MCP tools is a much better interface than the dashboard for "who spent what" questions.
Question: does the Cursor API expose per-user/per-model spend at fine enough granularity — and freshness — that you could drive enforcement from it, not just reporting? Context: we build AgentShield, a pre-call spend firewall for agents (stdlib Python, per-transaction rules — velocity checks, daily caps, per-call ceilings). The gap we keep seeing is that analytics tools tell you on Tuesday that Monday was expensive, while the expensive failure (a stuck retry loop, $50–$200 in ~10 minutes) needs data fresh enough to trip a rule mid-run.
If cursor-usage can expose near-real-time per-user burn, a rule engine could subscribe to it — genuinely curious what the API's data latency looks like. https://agentshield.fly.dev