# Best Tools to Track Claude Code, Cursor and Copilot Spend Canonical: https://www.pointfive.co/blog/how-to-track-coding-agent-spend-across-your-team Compare tools to track AI coding agent spend: native views for Claude Code, Cursor, Copilot, Codex and Devin, local trackers like ccusage, and team platforms. By: Gal Ben-David Published: 2026-10-09 [All articles](https://www.pointfive.co/blog) # Best Tools to Track Claude Code, Cursor, and Copilot Spend in 2026 Gal Ben-David[LinkedIn](https://www.linkedin.com/in/gal-ben-david/) Co-founder & CPO, PointFive October 9, 2026 9 min read If your engineers use Claude Code, Cursor, GitHub Copilot, Codex, or Devin Desktop, someone will eventually ask a simple question: what are we spending on coding agents, and who is spending it? The answer is harder than it sounds. Each tool bills differently, reports differently, and stops at its own boundary. This guide compares the tools that track coding agent spend, grouped into three kinds: - Native vendor views: the admin dashboards and APIs each coding agent provides. - Local trackers: free, open-source tools that read usage files on a developer's machine. - Team platforms: products that collect usage from every agent and attribute it across the organization. All details below come from each vendor's own documentation or repository, checked in October 2026. ## Why coding agent spend is hard to track Three things make this harder than tracking a cloud bill. Billing models do not match. Claude Code and Cursor combine seats with usage-based charges. GitHub Copilot [moved to usage-based billing](https://docs.github.com/en/copilot/concepts/billing/billing-for-enterprises) with GitHub AI Credits in June 2026. Codex usage draws on ChatGPT workspace credits, and Devin Desktop (formerly Windsurf) reports consumption in its own credits and ACUs. There is no shared unit. Many numbers are estimates. A developer on a subscription does not pay per token, so most per-session figures are tokens multiplied by public list prices. That is useful for comparison, but it is not what finance pays. Each view stops at its own tool. A Cursor dashboard knows nothing about Claude Code. If your developers use three agents, you have three partial answers. ## Native vendor views Start here. They are free and they are the source of truth for what each vendor bills. Claude Code. Team and Enterprise admins get a [Claude Code analytics dashboard](https://code.claude.com/docs/en/analytics) with activity, accepted lines, and, with the GitHub app, pull requests shipped with Claude Code. Per-user spend comes from a separate spend report. API organizations get a Console dashboard and the [Claude Code Analytics API](https://platform.claude.com/docs/en/build-with-claude/claude-code-analytics-api) , which returns daily per-user records with tokens and estimated cost by model. Claude Code can also [export OpenTelemetry metrics](https://code.claude.com/docs/en/monitoring-usage) , including token usage and estimated cost, to any observability backend. Cursor. The [team analytics page](https://cursor.com/docs/account/teams/analytics) focuses on adoption and accepted code, not spend. Spend lives in the [Admin API](https://cursor.com/docs/account/teams/admin-api) , which returns per-member spend for the billing cycle and per-request usage events with tokens and charged amounts. Teams can set monthly spend limits; per-member limits are an Enterprise feature. GitHub Copilot. The [usage metrics dashboard and API](https://docs.github.com/en/copilot/concepts/copilot-metrics) cover adoption, engagement, and acceptance, but the metrics API contains no cost data. Spend is managed through billing, where budgets can be set at the user, cost-center, and enterprise level. Codex. ChatGPT Business and Enterprise workspaces have [usage analytics](https://learn.chatgpt.com/docs/enterprise/usage-insights) showing credits and active users by use case, an aggregated daily [analytics API](https://learn.chatgpt.com/docs/enterprise/analytics-api) , and spend controls on workspace credits. Devin Desktop (formerly Windsurf). An [analytics API](https://docs.devin.ai/admin/billing/usage.md) reports consumption and accepted code, groupable by user, and Enterprise admins can cap usage per user. The gap: every one of these answers "what did this tool cost" well. None answers "what did AI-assisted development cost this team, across every tool it used." ## Local trackers for individual developers A wave of open-source tools reads the usage logs that coding agents write to disk and turns them into cost reports. They are free, private by default, and quick to install. They are built for one developer on one machine, and their cost figures are estimates based on public model prices. | Tool | What it does | Agents covered | License | | --- | --- | --- | --- | | [ccusage](https://github.com/ccusage/ccusage) | CLI reports of tokens and estimated cost by day, month, session, and 5-hour block | 18 agents, including Claude Code, Codex, OpenCode, Gemini CLI, and Copilot CLI | MIT | | [CodeBurn](https://github.com/getagentseal/codeburn) | Cost by tool, model, project, branch, task, and pull request, with optimization suggestions | 37 tools, including Claude Code, Codex, Cursor, Copilot, and Devin | MIT | | [agentsview](https://github.com/kenn-io/agentsview) | Local session search with a token and cost dashboard | 20+ agents, including Claude Code, Codex, and Cursor | MIT | | [Tokscale](https://github.com/junhoyeo/tokscale) | CLI and terminal UI for token and cost tracking, with an opt-in public leaderboard | 20+ clients, including Claude Code, Codex, and Gemini | MIT | | [OpenUsage](https://github.com/robinebers/openusage) | macOS menu bar app for plan limits, credits, and spend | Claude, Codex, Copilot, Cursor, Devin, and others | MIT | ccusage is the most widely used, with close to 19,000 GitHub stars, and CodeBurn has grown quickly since its 2026 launch. For a developer who wants to understand their own usage, either is a good start. What local trackers cannot do is the organizational part. Each developer sees their own numbers, there is no shared view, no attribution to teams, and no policy control. ## Team platforms Team platforms collect usage from every agent, normalize it, and attribute it across the organization. They differ in how they collect data and what they can do with it. [TokenShift](https://www.pointfive.co/products/tokenshift) by PointFive is built specifically for AI agents. It shows spend by developer, team, model, and session across Claude Code, Codex, Cursor, GitHub Copilot, Devin Desktop, and OpenCode, and it connects the cost of a task to its output, from merged pull requests to discarded code. Organizations can set model policies and spend caps by team and block specific tools. Analysis runs locally, where the agent runs; only derived metadata and analysis results reach PointFive's control plane, and prompt content does not. Beyond tracking, it adds optimization: context optimization reduces unnecessary context without changing the model, and Smart Routing is a separate, opt-in module. Vantage connects to [Cursor](https://docs.vantage.sh/connecting_cursor) through its Admin API, to Anthropic through its usage and cost API, and to [GitHub](https://docs.vantage.sh/connecting_github) for Copilot AI Credits, showing coding agent cost alongside cloud spend. Datadog offers an [agents console](https://docs.datadoghq.com/ai_agents_console/setup/) , in preview, covering Claude Code, Cursor, and GitHub Copilot, and Grafana Cloud has a [Claude Code integration](https://grafana.com/docs/grafana-cloud/monitor-infrastructure/integrations/integration-reference/integration-claude-code/) built on its OpenTelemetry metrics. These fit teams that already run those platforms and want agent usage next to the rest of their telemetry. SaaS-spend tools such as Torii and StackSpend track AI tool spend as part of a broader software or cloud spend view. They are useful for procurement and seat management, with less depth on how agents are used. ## Side-by-side comparison | | Native views | Local trackers | Team platforms | | --- | --- | --- | --- | | Who it is for | Admins of one tool | Individual developers | Engineering leaders, platform and FinOps teams | | Covers multiple agents | No | Yes, on one machine | Yes, across the organization | | Cost figures | Billed (in billing pages) or estimated | Estimated from list prices | Depends on the platform's data source | | Attribution to teams | Limited | No | Yes | | Policies and spend caps | Per tool | No | Varies; TokenShift sets them by team | | Cost | Included | Free | Paid | ## How to choose - One agent, small team: use the vendor's native view and set its spend limits. It is free and accurate. - Individual developers who want to see their own usage: install ccusage or CodeBurn. It takes minutes and builds good habits. - Several agents across many teams: you need a team platform. Choose by data source (billing APIs, telemetry, or local agent data), attribution depth, and whether you need controls such as policies and caps, not only reports. ## Track the cost of the work, not just the tool Whatever you pick, avoid stopping at "spend per tool" or "spend per seat." The useful question is what each completed piece of work cost. Two developers can spend the same amount, and one ships a merged change while the other's session ends in discarded code. Measuring [cost per successful task](https://www.pointfive.co/blog/llm-cost-optimization-cost-per-successful-task) shows the difference. For the wider picture, see [what FinOps for AI means](https://www.pointfive.co/blog/what-is-finops-for-ai) and our guide to [attributing coding agent costs to teams, projects, and outcomes](https://www.pointfive.co/guides/attribute-coding-agent-costs-teams-projects-outcomes) . ## Frequently asked questions What is the best tool to track Claude Code spend? For one developer, ccusage or CodeBurn give a free local view of tokens and estimated cost. For an organization, Anthropic's Claude Code analytics and Analytics API cover Claude Code alone, while team platforms such as TokenShift combine Claude Code with every other coding agent and attribute spend to teams. Is there a single dashboard for Copilot, Cursor, and Claude Code spend? Not from the vendors; each reports only its own tool. A cross-tool view needs a team platform that collects usage from every agent, such as TokenShift, or a manual consolidation of each vendor's exports. Are local trackers like ccusage accurate? They are accurate about tokens, because they read the agent's own logs. Their dollar figures are estimates based on public model prices, which can differ from what you pay on a subscription or a negotiated contract. Can I attribute coding agent costs to teams? Yes, but rarely with native views alone. Attribution needs a mapping from developers to teams and data from every agent they use, which is what team platforms are built to provide. ## Related reading [All articles](https://www.pointfive.co/blog) - ### [The fourth layer: why coding agents are about to become your biggest AI bill](https://www.pointfive.co/blog/the-fourth-layer-coding-agents-biggest-ai-bill) June 24, 2026 6 min read - ### [Coding agent cost management: a radically new practice, and a badly misunderstood one](https://www.pointfive.co/blog/coding-agent-cost-management-discipline) August 18, 2026 10 min read - ### [Token reduction is not cost reduction](https://www.pointfive.co/blog/token-reduction-is-not-cost-reduction) August 6, 2026 3 min read --- Source: the public page above. Product screenshots and illustrative interfaces are examples, not live customer data.