Claude Enterprise adds project tags to track AI spend by cost code

2 minutes ago

8 min read

Anthropic has added project tags to Claude Enterprise, a beta feature that lets organizations attach their own accounting codes to Claude projects and then see most of the spend generated inside those projects grouped by code. The change appeared in the Claude release notes on November 6, 2026, with a support article describing how tags are created, who can set them, which Claude products report spend against them, and how the same data can be pulled through Anthropic's Admin and Analytics APIs.

The codes are the ones finance teams already use: a work breakdown structure (WBS) code, an internal charge code, or a client-matter number in a law or consulting firm. A tag is the category (for example cost_code) and a value is the specific code (for example CC-1042). Each project can carry up to 16 tags, each tag holds one value per project, and spend can then be grouped by any tag in organization settings, exported as CSV, or retrieved from a usage and cost reporting endpoint.

The feature is limited to Claude Enterprise, is labelled beta, and is switched on automatically for eligible organizations, although it does nothing until an Owner creates the first tag. Several Enterprise configurations are excluded, including organizations with a HIPAA configuration, customer-managed encryption keys, zero data retention or data stored outside the US, so many regulated customers will not see it in this first release.

··········

WHAT CLAUDE ENTERPRISE PROJECT TAGS DO.

Project tags are designed for the chargeback work finance teams already do, mapping AI consumption to the codes their systems use to bill costs back to departments, clients or internal projects. The mapping happens at the project level: if a team runs its work inside a Claude project tagged cost_code = CC-1042, the spend generated in that project is reported under CC-1042.

Anthropic describes the scope carefully. Tags apply only to work that happens inside a project, and work outside a project receives no tag at all. The support article says tags let organizations see "most" spend by tag, and the exclusions explain why: several Claude surfaces do not yet report against projects, and usage covered by seat licences is not counted in the spend figures. Amounts are shown after discounts and before credits, so the totals are a management view of consumption and may not match invoices. Anthropic also warns that recent figures can still change until about a week after the end of the month.

The design keeps tag names fixed and values flexible. Tag names are lowercase with no spaces and cannot be renamed after creation. For each tag, an Owner chooses whether users may enter any value or only pick from a predefined list, which can be typed in one value per line or imported from a single-column CSV whose header is value. Values can be up to 256 characters and may contain spaces and commas, which suits client-matter strings. Claude does not validate values against an ERP or finance system, so a predefined list only restricts what users can select.

··········

WHICH CLAUDE PRODUCTS REPORT SPEND BY PROJECT TAG.

Coverage is the first thing a finance or IT team will need to check, because any usage outside the covered surfaces will appear as untagged. The support article lists the following.

........

Claude surface	Spend reported by project tag	Condition
Chat inside a project	Yes	Work must happen in the project
Unified chat and Cowork inside a project	Yes	Work must happen in the project
Cowork tasks on your computer	Yes	Claude Desktop 3.6556.1 or later, one project chosen for the task
Claude Code in terminal or desktop app	Yes	Project set in settings; Claude Code 2.1.305+ or Claude Desktop 3.6556.1+
Claude Code on the web	No	Not covered in the beta
Claude for Office and Claude in Chrome	No	Not covered in the beta
Claude Design, Artifacts and voice mode	No	Not covered in the beta
Claude in Slack and Microsoft Teams	No	Not covered in the beta
Claude API and Console	No	Not covered in the beta
Claude Code via Bedrock, Vertex AI, a gateway or an API key	No	Not covered in the beta

........

The practical consequence is that an organization whose developers mostly call Claude through the API, or run Claude Code through Amazon Bedrock or Google Cloud Vertex AI, will see little of that spend under project tags. The feature is aimed at Claude's own apps and at Claude Code sessions authenticated through the Enterprise plan.

··········

HOW TO SET UP PROJECT TAGS IN CLAUDE ENTERPRISE.

Only a Primary Owner, an Owner, or a custom role with Owner access can create tags and manage their values. The path is Organization settings, then Organization, then Project tags, where the Owner clicks "Create tag", names it, chooses between "Predefined values only" and "Any value", and optionally marks the tag as Required so that users must set it on new projects. More values can be added later with "Add values" or "Import CSV".

Setting a value on a project requires "Can edit" access to that project. Users open the project, go to Tags, click "Add tag", choose the tag and pick or enter the value. Tag names and values are visible to everyone in the organization, so Anthropic advises keeping sensitive information out of them, which is a point to raise with legal teams that use client names inside matter codes.

Claude Code has no project picker, so the project is assigned through a settings file. Anthropic documents a single key, attributionProject, which can sit in .claude/settings.json inside a repository or in ~/.claude/settings.json to apply to every repository on a machine. The project ID is the last segment of the project's URL in Claude.

{
  "attributionProject": "PROJECT_ID"
}

If no project is set, Claude Code usage is simply not tagged and no warning appears, so teams that rely on the reports will want to distribute the settings file through their standard developer environment setup. The JSON file must remain valid, and Claude Code may ask the user to trust the folder before it applies the setting.

........

Task	Who can do it
Create tags and manage values	Primary Owner, Owner, or custom role with Owner access
Set a project's tag values	Anyone with "Can edit" access to the project
View spend by tag	Primary Owner, Owner, or custom role with Analytics view and Billing view
Create API keys for the Admin API	Primary Owner

........

··········

READING CLAUDE ENTERPRISE SPEND BY COST CODE.

Spend by tag appears under Organization settings, then Usage, then Spend by project. Administrators use "Group by" to select the tag and narrow the view with the "Period" and "Product" menus; results can be downloaded with "Export CSV". Data runs through the end of the previous day in UTC, so the report is a daily view rather than a live meter.

Two labels in the report deserve attention during a rollout. Spend outside any project appears as "Not in a project", which also includes Claude Code sessions with no project configured. Spend in a project that lacks a value for the selected tag appears as "No tag value". Watching those two lines shrink over the first weeks is the simplest way to measure adoption. Private projects are listed as "Private project" on this page and in the CSV, while the Analytics dashboard, the Analytics API and the Compliance API show their real names.

Timing rules also affect accuracy. A new tag or a change to a tag takes about two hours to apply, and usage in that window may stay under the previous value or under no value. Spend before a tag takes effect never counts toward it, and changing a tag does not rewrite past spend, so Anthropic recommends tagging a project before work begins. Reporting starts on the date the feature became available to the organization, which means earlier months cannot be reconstructed by tagging old projects.

··········

PROJECT TAGS IN THE CLAUDE ADMIN AND ANALYTICS APIS.

Both APIs involved are in beta, and Anthropic notes that paths and names may change. Through the Admin API, a key with the read:project_tags and write:project_tags scopes can list tags, add values and set a project's tags with a POST to /v1/organizations/apps/projects/{claude_proj_ID}/tags, using the beta header anthropic-beta: ce-project-tags-2026-10-09. That endpoint expects the internal claude_proj_ identifier rather than the ID in the project URL, and it cannot create tags, archive values or create projects, which remain tasks for the web console.

To pull spend into a finance system, an organization enables the Analytics API and uses a key with the read:analytics scope. The documented call is GET /v1/organizations/analytics/cost with daily buckets and a grouping on the tag, sent with the beta header anthropic-beta: analytics-usage-cost-2026-08-24. The request below is assembled only from the parameters in Anthropic's documentation and is shown for illustration.

GET /v1/organizations/analytics/cost?time_bucket_width=1d&group_by[]=tags.claude_project[cost_code]
anthropic-beta: analytics-usage-cost-2026-08-24

Each call covers up to 31 days of daily buckets and returns up to 1,000 rows; when has_more is true the client must keep paging or the totals will be incomplete. Amounts come back in cents, can include fractions of a cent, and are net of discounts. Because the tag filter text travels in the URL, Anthropic cautions that it may appear in request logs, another reason to keep confidential client names out of tag values.

··········

LIMITS OF THE CLAUDE PROJECT TAGS BETA.

The beta leaves several controls for later. Anyone who can edit a project can change or clear its tags unless the tag is required, and tags cannot be locked. Required tags apply only to new projects created on the web; a required value can be changed there but not removed, although the Admin API can still remove it. Tags cannot yet be placed on a single chat or session, spend cannot be split across several values of the same tag, tags and values cannot be renamed, and there are no spend limits by tag. Budget controls therefore remain a separate tool from tag reporting; the pooled group budgets Anthropic added for Enterprise admins in October are covered in our report on Claude Enterprise smart reports and pooled group budgets.

Eligibility is the other constraint. Besides the HIPAA, customer-managed keys, zero data retention, minimal retention and non-US data storage exclusions, organizations that route Claude Desktop through their own model provider are also left out. Anthropic says turning project tags off requires contacting the account team. For the wider picture of what each Claude plan includes, see our guide to Claude plans and usage limits, and for the security and deployment side of the Enterprise plan, our explainer on Claude Enterprise SSO, SCIM, audit logs and EDR.

··········

CLAUDE ENTERPRISE PROJECT TAGS: CONFIRMED FACTS AND OPEN QUESTIONS.

Anthropic's documentation establishes the mechanics in detail: Enterprise-only beta, up to 16 tags per project with one value each, Owner-managed tags with optional predefined lists and a Required switch, daily reporting with CSV export, an Admin API to set values and an Analytics API endpoint that groups cost by tag. It also establishes the gaps, starting with the absence of API, Console, Claude Code on the web, Office, Chrome, Slack and Teams usage from the tagged figures, and the exclusion of HIPAA, zero-retention and non-US-storage customers.

Anthropic has not said when the uncovered surfaces will report by project, when the regulated configurations will become eligible, or whether spend limits by tag will follow the reporting. For finance teams the current release already allows a monthly chargeback of Claude chat, Cowork and plan-authenticated Claude Code spend to existing cost codes, provided projects are tagged before work starts, developers set attributionProject on their machines, and the reports are read as management figures rather than invoice totals.

··········

·····

DATA STUDIOS

·····

datastudios.org