cursorscope

Cursor generates a lot of signal. Sessions, prompts, tool calls, model switches, compactions, subagents. None of it goes anywhere by default. cursorscope fixes that.

It's a small Node.js service that intercepts Cursor's hook events and exports them as OpenTelemetry traces, metrics, and logs to any OTLP-compatible backend. Last9, Grafana Cloud, Honeycomb, Datadog, Jaeger, a local Collector — whatever you already run. You get full observability into how your team uses Cursor: which models they reach for, how long agent loops run, where tools fail, how token budgets are spent.

How it works

Cursor fires lifecycle hooks (sessionStart, beforeSubmitPrompt, preToolUse, etc.) as JSON on stdin to a shell script. That script hands the payload to a Node forwarder, which posts it to a local HTTP ingestor. The ingestor maps hook events onto OTel spans — correlating by conversation ID and generation ID so tool calls nest correctly inside their parent prompt spans — and exports everything via OTLP.

The result is a proper distributed trace for every agent interaction: session span at the root, prompt spans as children, tool call spans underneath those, subagent spans branching off when Cursor fans out work.

Backends

cursorscope speaks OTLP HTTP. Point it anywhere:

OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=https://your-backend/v1/traces
OTEL_EXPORTER_OTLP_METRICS_ENDPOINT=https://your-backend/v1/metrics
OTEL_EXPORTER_OTLP_LOGS_ENDPOINT=https://your-backend/v1/logs
OTEL_EXPORTER_OTLP_HEADERS=Authorization=Bearer <your-token>

| Backend | Notes | |---------|-------| | Last9 | Guided setup via npx @last9/cursorscope | | Grafana Cloud | OTLP endpoint + basic auth from your stack settings | | Honeycomb | api.honeycomb.io:443, header x-honeycomb-team=<api-key> | | Datadog | OTLP Agent endpoint localhost:4318 via local Datadog Agent | | Jaeger / Tempo | Local collector or direct OTLP endpoint | | Local Collector | docker compose up -d — see With an OTel Collector |

Setup

Last9

One command installs to ~/.cursorscope, writes .env, registers global Cursor hooks, and starts the ingestor:

npx @last9/cursorscope

Interactive setup opens Last9 → OpenTelemetry integration so you can copy the OTLP endpoint and Basic auth token.

Note: The OTLP auth token is only visible to Last9 organization admins. If you're not an admin, ask yours to copy it from the integration page.

Non-interactive:

npx @last9/cursorscope setup --last9 \
  --otlp-base <your-last9-otlp-endpoint> \
  --auth-token "$LAST9_OTLP_TOKEN"

Any other OTLP backend

npx @last9/cursorscope setup

This runs the same guided flow without the Last9 preset — you supply your own endpoints and auth headers.

After setup

Restart Cursor and send one Agent message. The ingestor starts automatically in the background.

If you ever need to intervene:

npx @last9/cursorscope status        # health check — is the ingestor running?
npx @last9/cursorscope start         # start it manually if auto-start failed
npx @last9/cursorscope hooks install # re-register Cursor hooks after a Cursor update

Manual setup

For contributors or anyone who wants full control.

Clone and install:

git clone https://github.com/last9/cursorscope.git
cd cursorscope
npm install

Copy the env file and fill in your OTLP destination:

# run from repo root
cp .env.example .env

The minimum to export to a remote backend:

OTEL_EXPORTER_OTLP_ENDPOINT=https://your-otlp-endpoint
OTEL_EXPORTER_OTLP_HEADERS=Authorization=Basic <base64-credentials>
OTEL_EXPORTER_OTLP_METRICS_TEMPORALITY_PREFERENCE=cumulative

Auto-start (recommended)

Install global hooks once — merges into ~/.cursor/hooks.json without touching your existing hooks (rtk, custom scripts, etc.). A timestamped backup is created first.

# run from repo root
npm run install:global-hooks

Restart Cursor and open a new chat. The ingestor starts in the background automatically. Logs at ~/.cursor/cursorscope.log.

# run from repo root
npm run stop                    # stop the background ingestor
npm run uninstall:global-hooks  # remove only cursorscope hook entries

Set CURSORSCOPE_AUTO_START=false in .env if you prefer npm start by hand.

Register all hook types (optional): CURSORSCOPE_HOOK_EVENTS=all npm run install:global-hooks

Per-project hooks

Point Cursor at .cursor/hooks.json in this repo. Hooks call .cursor/hooks/forward.sh, which auto-starts the ingestor and forwards events.

Manual start

# run from repo root
npm start

For a central install, set CURSORSCOPE_HOME and CURSOR_HOOK_ENDPOINT to your running service URL.

With an OTel Collector

To fan out to multiple backends or pre-process telemetry, run the included Collector:

docker compose up -d

This starts otel/opentelemetry-collector-contrib on localhost:4317 (gRPC) and localhost:4318 (HTTP). The local config (otel/collector.local.yaml) sends everything to the debug exporter so you can see what's being emitted. Swap in otel/collector.last9.yaml or write your own to route to a real backend.

Cursor Admin API polling

Cursor Business accounts can unlock team-level daily usage metrics — requests, tokens, model breakdowns — exported as OTel gauges:

ENABLE_CURSOR_API_POLLING=true
CURSOR_ADMIN_API_KEY=<your-key>
CURSOR_TEAM_ID=<your-team-id>
CURSOR_API_POLL_INTERVAL_MS=300000

Privacy

User prompts are redacted by default — only prompt length is recorded. Set CURSOR_LOG_USER_PROMPTS=true to include prompt text in spans and logs. API keys and bearer tokens are scrubbed automatically regardless. Set CURSOR_MASK_USER_EMAIL=true to mask cursor.user.email before it leaves your machine.

What you get

Traces — one root span per session, one span per prompt/generation, tool call spans as children with duration and success/failure, subagent spans branching from parent conversations.

Metrics — cursor_hook_events_total, cursor_session_total, cursor_prompt_total, cursor_tool_executions_total, gen_ai.client.operation.duration, gen_ai.client.token.usage, cursor_api_metric_value (Admin API polling).

Logs — one structured log record per hook event with conversation ID, model, and hook-specific fields.

Attribute names follow the OTel GenAI semantic conventions where applicable.

Debug endpoints

GET  /healthz               — liveness check
GET  /debug/otel-config     — show resolved OTLP endpoints and config
POST /debug/emit-and-flush  — emit a test trace/metric/log and force flush
GET  /debug/otlp-probe      — probe connectivity to all configured OTLP endpoints

Requirements

Node.js 20 or later. No build step.