Claude Console can help you audit API activity, but its documented dashboard is narrower than a project-management or workflow monitor. Open Claude Console → Usage, export the CSV, and investigate usage by API key and model. That export is useful for finding applications still calling deprecated models. It does not, according to the cited documentation, provide native project-name, workflow-status, user, real-time, spend-alert, or task-level views.
If you need developer or team attribution, budgets, rate limits, and centralized audit logs, put a separately evaluated LLM gateway in front of your applications. Treat those as gateway capabilities, not features of the Claude Console Usage page.
As an Amazon Associate I earn from qualifying purchases.
What Claude’s dashboard actually monitors
The relevant Claude dashboard for this task is the API Console’s Usage page. Anthropic documents an Export action that downloads a CSV of API activity. The documented breakdown is by API key and model; Anthropic’s stated use case is identifying applications that still use deprecated models.
Free tools Windows power users keep installed
One-click scans. No signup required.
That distinction matters when someone asks to “monitor Claude projects and workflows.” An API key can represent an application or an environment, and a model identifies which Claude endpoint that key called. Those fields are useful evidence for an audit, but they are not the same as a project name, workflow run, human owner, ticket, or live job status.
#1 Best Overall
The export documentation does not establish project-level grouping, workflow tracking, user attribution, real-time refresh, spend alerts, or status monitoring. Do not promise those views based on the Usage page alone. If your organization needs them, you must add an attribution layer in your own systems or evaluate a separate gateway.
Anthropic’s model-deprecation guidance states: “Review the downloaded CSV to see usage broken down by API key and model.” Read the current instructions at Anthropic’s model-deprecations documentation.
How do I monitor Claude usage by project?
You cannot assume that Claude Console will display a project column or a workflow hierarchy. The reliable native method is to use API keys as boundaries for your own projects or environments, then interpret the export with an internal key registry.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Use a deliberate key-to-project convention
Before exporting data, record which key belongs to which application, environment, and owner in a controlled inventory. For example, a key may be assigned to a production summarization service, while another belongs to a staging evaluation job. The Console export can then tell you which key and model generated activity; your inventory supplies the project meaning that the export itself does not document.
- Give every application or environment a distinct API key where your access controls and operating practices allow it.
- Keep the key’s owner, application, environment, and rotation contact in an internal record rather than embedding secrets in a spreadsheet.
- Document the model expected for each service so an unexpected model in the CSV is easy to investigate.
- Rotate or revoke credentials according to your security policy; never paste secret key values into a shared report.
Understand what this method can and cannot prove
A key-level match can show that activity is associated with a credential assigned to a project. It cannot prove which individual user initiated a request, which workflow step ran, or whether a request is still in progress. A shared key also collapses all callers into one line of attribution. If those distinctions matter, separate the credentials or add application-side request metadata and logging.
Where can I see Claude API usage?
- Sign in to Claude Console. Use the account and organization that owns the API activity you need to review.
- Open the Usage page. This is the documented location for the API usage export.
- Select Export. Save the downloaded CSV in an access-controlled location. Treat it as operational data because it identifies keys and models associated with usage.
- Open the CSV with your approved analysis tool. Preserve the original file and work from a copy so the downloaded evidence remains intact.
- Filter by API key and model. Start with the key assigned to the project or environment under review, then inspect the model values associated with it.
- Compare unexpected model values with your approved model list. A deprecated model, an unplanned migration, or a test key can explain a change in the export.
- Record the finding and owner. Link the key and model observation to your internal project record, incident, or migration ticket.
The export is a point-in-time file, not a continuously updating dashboard. Repeat the export on a schedule appropriate to your change and deprecation risk, and label each file with the retrieval date and the account or organization reviewed.
Rank #2
- Used Book in Good Condition
How to turn the export into a project review
A small amount of process around the CSV can make a key-and-model report useful without misrepresenting it as a native workflow dashboard.
Step 1: Preserve the source file
Store the original download with restricted access and a clear date. Do not overwrite it after filtering. A read-only archive lets you explain later which export supported a migration or incident decision.
Step 2: Join keys to your project register
Use the API key identifier represented in the export to look up the owning application, environment, and technical contact in your internal register. If a key is missing from the register, classify the row as an attribution gap and investigate rather than guessing its project.
Step 3: Group by model within each key
For each assigned key, list the models present in the CSV. Compare that list with the models your application is expected to call. This highlights a service that has not migrated, a configuration that changed, or a credential being used by an unrecorded application.
Step 4: Escalate ambiguous or shared credentials
If one key serves several projects, the CSV cannot separate them. Create a remediation item to split the key or add application-side attribution. Do not allocate usage to a project solely because that project is the most likely caller.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Step 5: Publish a narrow, traceable report
A useful report can contain the export date, key identifier or an approved redacted form, mapped project and environment, observed models, expected model, owner, and follow-up status. Keep the report’s wording precise: “activity associated with key X” is supported; “workflow Y used model X at 10:32” is not supported by the documented export alone.
How to audit deprecated-model usage
Model migration is the clearest documented reason to use the Usage export. Anthropic’s guidance specifically positions the CSV as a way to identify applications that still use deprecated models.
- Read the current deprecation notices. Start with Anthropic’s model-deprecations page and make a list of models that require action for your account or application estate.
- Export current usage. Follow the Usage-page procedure and retain the original CSV.
- Search the model field for deprecated names. Use exact model identifiers from the current documentation; do not rely on display-name guesses.
- Map each matching key to an application. Use your internal key register. An unregistered key should become an ownership investigation, not an automatic migration assignment.
- Open a migration task for each affected application. Record the current model, intended replacement, owner, test plan, and target date in your normal change system.
- Update the application configuration and validate it. The Console export identifies the caller and model; your application tests must establish whether the replacement behaves correctly for that workflow.
- Export again after the change. A later export can show whether the affected key still appears with the deprecated model. It does not, by itself, prove that every request path or workflow branch has been exercised.
Keep the model-deprecation page bookmarked because model availability and retirement information can change. The CSV is evidence of observed API activity for the export period, not a guarantee about future calls.
What the Usage page does not provide
Several commonly requested dashboard functions are not established by the cited Console documentation:
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problems- Project-name filtering: the documented CSV breakdown is by API key and model, not by a native project field.
- Workflow or task status: the export does not document running, queued, failed, or completed workflow states.
- Individual-user attribution: a key/model breakdown does not identify the person who initiated a request.
- Real-time monitoring: an exported CSV is a downloaded review artifact, not a live stream.
- Spend alerts and budget enforcement: those controls are not promised for the Usage export.
- Request-level business context: the cited guidance does not promise ticket IDs, customer IDs, job names, prompts, or response outcomes.
These are boundaries, not reasons to discard the export. Key-and-model evidence is well suited to credential inventories, model-deprecation audits, and investigations into which application boundary is generating activity.
When a team needs broader workflow attribution
Anthropic’s Claude Code documentation describes a third-party LLM gateway as a centralized layer. The listed gateway capabilities include attributing usage by developer or team, enforcing budgets and rate limits, maintaining audit logs, and switching providers through gateway configuration. Those capabilities belong to the gateway layer; they should not be presented as native Claude Console Usage features.
| Monitoring approach | Attribution documented | Review or control model | Operational responsibility |
|---|---|---|---|
| Claude Console Usage export | API key and model | Download and review a CSV; project and workflow grouping are not documented | Maintain key ownership and interpret the file inside your organization |
| Third-party LLM gateway | Developer or team attribution is listed as a gateway capability | Budgets, rate limits, audit logging, and provider switching are listed as gateway capabilities | Operate, secure, configure, and evaluate the additional gateway layer |
Choose the Console export when key-and-model auditing answers the question. Consider a gateway only when centralized attribution or policy enforcement justifies the extra component and its security review.
How to evaluate a gateway without confusing it with Claude Console
Define the attribution you actually need
Write down whether you need an application key, developer, team, department, environment, customer, or workflow-run identifier. The Console export addresses the first two fields—API key and model—at the documented level. A gateway may add developer or team attribution, but you still need to verify how it obtains that identity and whether it fits your authentication model.
Specify policy requirements
Decide whether you need budgets, rate limits, audit logs, provider switching, or all four. Ask how policies are configured, where logs are stored, how long they are retained, and how administrators investigate an exception. Do not infer implementation details from a feature list.
Perform a security and reliability review
A gateway sits between applications and the model provider. Review its secret handling, access controls, network path, failure behavior, logging of sensitive data, update process, and recovery plan. Test what happens when the gateway is unavailable or misconfigured, and decide whether applications have an approved fallback.
Read Anthropic’s qualification
Anthropic identifies LiteLLM as third-party and says it does not endorse, maintain, or audit that product’s security or functionality. Treat that statement as a reminder to conduct your own product, code, vendor, and security assessment. The gateway documentation is available at Anthropic’s “Other LLM gateways” page.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.A practical operating cadence
Use a cadence that matches your risk rather than assuming the Usage page is an always-on monitor.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →- At model-deprecation announcements: export immediately, search for affected model identifiers, and assign each matching key to an owner.
- During migrations: export before the change, after the change, and again after the application’s normal traffic has exercised relevant paths.
- For new applications: register the key, intended model, environment, and owner before production traffic begins.
- For unexplained activity: export, identify the key/model pair, check recent deployments and key ownership, and rotate or revoke credentials if compromise is suspected.
- For periodic governance: archive exports with retrieval dates and review unregistered keys, shared keys, and models outside the approved list.
This cadence creates an auditable trail while keeping the scope honest: the Console supplies API activity evidence, and your project register, deployment records, and incident system supply the surrounding context.
Best Value
Troubleshooting common monitoring questions
“I exported the CSV, but I cannot find a project column.”
That is consistent with the documented breakdown. Use the API key to join the row to your own project register. If one key serves multiple projects, the file cannot reliably separate their activity; split the credential boundary or add an attribution layer.
“A key appears with a model we did not expect.”
Verify the exact model identifier, check recent configuration changes, and identify every application that can access the key. Compare the value with the current deprecation information. Do not conclude that a particular user or workflow made the call unless another system records that link.
“Can I use the export for live incident response?”
Use it as an audit artifact, not as a documented real-time feed. For live response, pair the export with your application logs, deployment history, and credential controls, or evaluate a gateway that meets your real-time policy requirements.
Recommended Free Tools
“Can the export tell me whether a workflow succeeded?”
Not from the documented key-and-model fields alone. Success, failure, queue state, and business outcome belong in the workflow system or application telemetry that initiated the API call.
“Should I add a gateway just to get project names?”
First determine whether separate API keys and a maintained key register solve the attribution problem. A gateway adds a separate operational and security responsibility; adopt one when its centralized developer/team attribution or policy controls are necessary, not because the Console export is mislabeled as a workflow dashboard.
Related setup and reference documentation
For Claude Code configuration context, consult Anthropic’s Advanced setup documentation. For command-line details, see the CLI reference. These pages can help with surrounding setup, but they do not change what the Usage export documents: API activity broken down by key and model.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
