ClaudeSee what your Claude agents cost.
Claude Code, Cowork and Office usage streams in over OpenTelemetry, priced per model and engineer, in the same ledger as everything else.
- Read-only access
- 14-day free trial
- No credit card required
- 5 min
- setup, per provider
- 90 days
- available history
- Same-day
- anomaly alerts
Spend vs Budget
Forecast $45,000 this month
See what is driving Claude spend.
Price the tools that have no billing API.
Claude Code, Cowork and Office agents emit OpenTelemetry usage straight to a StackSpend endpoint — nothing to install, no credentials stored. Usage is priced per model, user and session as a token-based estimate, and labelled as one.
How it worksEvery model you run, in one lens.
Usage by base model, project and user, in tokens and in API-equivalent value. Estimated and billed usage stay separate, so the numbers never double-count and never pretend to be your invoice.
Business plan
How it worksClaude · token mixIllustrative example
1,000,000 tokens
- Claude Sonnet 5.5550,000
- Claude Haiku 4.5300,000
- Claude Opus 5.5150,000
600k input · 250k cached input · 150k output. API value is an estimate, not your invoice.
One message each morning. Nobody opens a billing portal.
Team plan and above
How it worksProduct examples are illustrative. Usage estimates and provider-reported costs are separate measures; availability varies by connected source.
Explore the model viewWhy is Claude spend hard to control?
- Claude usage can happen across developer tools, Cowork, and Office agents. Without one collector, teams cannot see where usage is coming from or which users and sessions are driving cost.
- Claude OpenTelemetry includes many event types. Cost signals, token usage, tool events, and logs need normalization and deduplication before they are useful for finance or operations.
- Claude sits alongside Anthropic API, OpenAI, Cursor, and cloud providers. Reviewing each source separately makes total AI spend hard to explain.
Know what you are connecting.
The coverage
- StackSpend gives each Claude provider connection a scoped ingest key and copy-ready OpenTelemetry settings for Claude Code, Cowork, and Office agents.
- Claude telemetry is normalized into daily cost rows with user, model, session, and source attribution so teams can separate Claude Code, Cowork, and Office agent usage.
- Daily alerts, anomaly detection, budgets, and pace-to-forecast help teams spot unexpected Claude usage before the month closes.
The source
Billing and usage from your connected providers. Credential types and permission controls vary by provider.
Provider connection guidesThe limits
Provider reporting and scheduled sync determine freshness. Available history and attribution vary by source; review the setup guide for coverage. Alerts notify your team; they do not block requests or enforce a spending cap.
What we track
- Claude Code OpenTelemetry
- Claude Cowork OpenTelemetry
- Claude Office agents OpenTelemetry
- Estimated cost by model
- User and session attribution
- Daily spend and anomaly detection
- Forward-looking telemetry from setup time, not 90-day backfill
Who should use StackSpend for Claude?
- Product and engineering teams that need model-level visibility before AI bills surprise them.
- Buyers consolidating OpenAI, Anthropic, Claude, Cursor, or open-model spend into one operating view.
- Teams that need alerts and forecasting, not just retrospective usage dashboards.
Evaluation checklist
- 01
Start a trial
Open a StackSpend workspace and add Claude as a provider connection.
- 02
Copy the Claude setup
Generate the provider ingest key and paste the managed settings or OTLP fields into Claude admin.
- 03
Send a test event
Start a new Claude session, run a small task, and confirm Last Logs updates on the provider page.
StackSpend alongside Claude admin monitoring and Anthropic billing views.
Native tools provide provider-specific reporting and controls. StackSpend adds a shared monitoring workflow across connected sources.
Claude admin monitoring and Anthropic billing views
- Telemetry setup is separate from cost reporting and needs collector configuration
- Raw OpenTelemetry events include noisy non-cost logs that are hard to read directly
- No unified view with OpenAI, Cursor, cloud, and other AI providers
- Estimated Claude tool usage is not automatically translated into daily budget signals
StackSpend
- Copy-ready Claude Code managed settings and Cowork/Office OpenTelemetry fields
- Normalized estimated cost rows with source tagging and deduplication
- User, model, and session breakdowns in the same dashboard as other providers
- Daily alerts, anomaly detection, and budget pacing for Claude usage
What do you get when you connect Claude?
- Access model
- OpenTelemetry ingest with a scoped usage:write provider key.
- Visibility
- Estimated cost by Claude source, user, model, and session.
- Coverage
- Claude Code, Cowork, and Office agents in one provider connection.
- Signals
- Daily alerts, anomaly detection, and budget tracking for Claude usage.
Check the details before connecting.
Review connection permissions
Read the provider setup guides before sharing credentials.
Provider setup guidesSee the security details
How credentials, tenant isolation and data handling work.
Security and data handlingTalk to the team
Ask about your stack or requirements before connecting.
Contact StackSpendAbout Andrew DayClaude Cost Monitoring, answered
What does StackSpend cover for Claude?
- Tracks Claude Code, Cowork, and Office agent telemetry through OTLP ingest.
- Separates estimated cost-bearing events from noisy logs and non-cost usage signals.
- Uses a scoped provider ingest key rather than Anthropic API billing credentials.
What causes Claude costs to spike?
- A managed Claude Code rollout expands to more developers without a matching daily usage signal
- Cowork or Office agent sessions start sending more Opus requests than expected
- A long-running Claude session produces repeated high-cost model calls during a refactor
- Telemetry includes both logs and metrics and needs deduplication before costs are trusted
- Internal Claude workflows became part of daily engineering or support operations.
- Long-context processing repeated across documents or customers without caching.
- Agent loops and tool calls increased requests per task.
- Model defaults or routing rules shifted toward more expensive usage.
What is StackSpend for Claude Cost Monitoring?
StackSpend receives Claude OpenTelemetry from Claude Code, Cowork, and Office agents. See estimated daily cost, user-level attribution, model breakdowns, anomaly detection, and budget tracking alongside your other AI and cloud providers. Claude telemetry is a token-based estimate, not an official Anthropic invoice: it is derived from usage events and does not factor in subscription (plan) pricing, so treat it as a directional signal rather than a billed amount. Historical 90-day backfill is not available — telemetry is captured from setup time forward.
How do I monitor Claude Code, Cowork, and Office agent costs?
StackSpend receives OpenTelemetry from Claude Code, Cowork, and Office agents through a scoped ingest key, then normalizes it into daily cost rows with user, model, session, and source attribution. This is a token-based estimate derived from usage events — a directional signal, not an official Anthropic invoice, and it does not factor in subscription pricing. Telemetry is captured from setup forward; 90-day backfill is not available.
How do I connect Claude telemetry to StackSpend?
Create a Claude provider connection in StackSpend to get a scoped ingest key and copy-ready OpenTelemetry settings for Claude Code, Cowork, and Office agents. Point your collector at those settings and StackSpend deduplicates and normalizes the events into daily cost. The ingest is receive-only — StackSpend reads telemetry and never writes back to your Claude environment.
How do I see which user or model is driving Claude usage?
StackSpend attributes Claude telemetry by user, session, model, and source, so you can separate Claude Code, Cowork, and Office agent usage and see who and which model is driving estimated cost. When a managed rollout expands or a session sends more Opus requests than expected, the daily view shows the shift instead of blending it into one total.
Why did my estimated Claude cost jump?
Estimated Claude cost usually jumps when a rollout adds developers, when Cowork or Office agent sessions send more Opus requests, or when a long-running refactor produces repeated high-cost calls. StackSpend runs anomaly detection against your baseline and flags the surge the same day with the likely driver, so you catch it before the month closes rather than after.
Can I see Claude usage alongside my OpenAI, Cursor, and cloud spend?
Yes. StackSpend puts estimated Claude cost in one view with Anthropic API, OpenAI, Cursor, Hugging Face, Grok, and cloud providers like AWS, GCP, Azure, Snowflake, and Vercel. Instead of reviewing each source separately, you get a single AI-and-cloud total with daily alerts, budgets, and pace-to-forecast across everything you have connected.
Why are my Claude costs higher than expected?
Usually internal Claude Code or agent workflows becoming daily operations, long-context processing without caching, agent loops increasing requests per task, or model routing shifting to pricier usage. Compare usage by workflow and model to find the driver.
How do I track what is driving Claude usage cost?
StackSpend ingests Claude OpenTelemetry from Claude Code, Cowork, and Office agents and breaks estimated cost down by source, user, model, and session — then flags anomalies the day a workflow starts spending unexpectedly.
Is Claude telemetry cost the same as my Anthropic invoice?
No. Claude OpenTelemetry cost fields are estimates. Subscription coverage, credits, or enterprise terms may differ from the telemetry estimate — use the Anthropic provider for API billing and the Claude provider for Claude product usage.
For the current provider catalogue, see supported integrations.
Tomorrow morning: your Claude number, in Slack.
Connect Claude today and follow spend, budgets and alerts in one place. Review provider permissions before connecting.