# Token Meter > Open-source, local-first AI coding agent usage and cost dashboard for Claude Code, Codex, Cursor, OpenCode, Kiro, and Pi. - **Type:** MCP server - **Install:** `agentstack add mcp-splunk-token-meter` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [splunk](https://agentstack.voostack.com/s/splunk) - **Installs:** 0 - **Category:** [AI & ML](https://agentstack.voostack.com/c/ai-and-ml) - **Latest version:** 0.1.0 - **License:** MIT - **Upstream author:** [splunk](https://github.com/splunk) - **Source:** https://github.com/splunk/token-meter - **Website:** https://splunk.github.io/token-meter/ ## Install ```sh agentstack add mcp-splunk-token-meter ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About 📝 Launch blog · 🌐 Website · 📚 Learn Tokenomics Token Meter is an open-source, local-first usage and cost dashboard for AI coding agents. It turns session evidence from Claude, Codex, Cursor, OpenCode, Kiro, and Pi into one view of token usage, estimated cost, context pressure, time, tools, and execution—so you can decide whether to continue, intervene, compare, or investigate a run. Python standard library only. No API keys for trace analysis. No Token Meter analytics or telemetry leaves your machine. ## Quick Start ### macOS or Linux ```bash git clone https://github.com/splunk/token-meter.git ./token-meter/scripts/install ``` The installer stages a stable per-user runtime, starts the local server and native companion, and configures automatic startup. For a browser-dashboard-only installation without the macOS menu-bar or Linux tray companion, use `./token-meter/scripts/install --backend-only`. This mode does not require the Swift toolchain or Linux GTK/AppIndicator packages, and it is preserved by automatic updates. ### Windows > **Beta:** The Windows extension is still in beta. ```powershell powershell.exe -NoLogo -NoProfile -ExecutionPolicy Bypass -Command '$p=Join-Path $env:TEMP "token-meter-bootstrap.ps1"; try { Invoke-WebRequest -UseBasicParsing "https://raw.githubusercontent.com/splunk/token-meter/main/scripts/bootstrap-windows.ps1" -OutFile $p; & $p } finally { Remove-Item -LiteralPath $p -Force -ErrorAction SilentlyContinue }' ``` The bootstrap uses WinGet from Microsoft App Installer to install missing Git and Python. It then stages the beta extension without administrator access. From an existing checkout, rerun `.\scripts\install-windows.cmd`. For a browser-dashboard-only installation without the Windows notification-area companion, add `-BackendOnly` to the downloaded bootstrap invocation (`& $p -BackendOnly`) or run `.\scripts\install-windows.cmd -BackendOnly` from an existing checkout. Automatic updates preserve this mode. ### Open Token Meter Open [http://127.0.0.1:8722](http://127.0.0.1:8722), start a normal agent run, and choose it from **Sessions**. For requirements, development startup, updates, uninstall commands, and troubleshooting, see the [User guide](specs/USER_GUIDE.md). **Windows beta uninstall:** `powershell.exe -NoLogo -NoProfile -ExecutionPolicy Bypass -File "$env:LOCALAPPDATA\Token Meter\runtime\scripts\uninstall-windows.ps1"` ## What You Can Do | Goal | Token Meter helps you | | --- | --- | | **Understand a live run** | Follow estimated cost, tokens, context pressure, wait, output pace, tool calls, execution evidence, and session-budget alerts. | | **Review history and spend** | Find expensive or slow work across sessions, projects, runtimes, platforms, and calendar ranges. | | **Compare models and execution** | Compare input, output, pace, wait, and workload shape without presenting weak matches as meaningful results. | | **Investigate tools and skills** | Find high-output, failing, repeated, unobserved, or deferred capabilities while keeping incomplete evidence explicit. | | **Manage usage** | Check provider-reported limits, allocate a monthly budget, receive threshold notifications, and let Codex or Claude query bounded evidence through the local MCP. | ## Coverage **Runtimes:** Claude Code and Desktop Agent/Cowork, Codex CLI and desktop, Cursor Agent/Composer, OpenCode, Kiro, and Pi. | Platform | Status | Experience | | --- | --- | --- | | **macOS** | Supported | Browser dashboard and native menu-bar companion | | **Linux** | Supported | Browser dashboard and AppIndicator tray companion | | **Windows** | **Beta** | Browser dashboard and notification-area extension | Evidence varies by runtime and client version. Missing values remain unavailable instead of appearing as a misleading zero. Token Meter works when the agent keeps session evidence on your machine in a supported local store. Sessions that exist only in a cloud-hosted service may not be available to Token Meter. ## First Five Minutes 1. Open **Sessions → Current sessions** and select an active run. 2. Under **Run**, check cost, context pressure, Output/$, and Reasoning ratio. Add a session budget if the run needs an attention limit. 3. After more sessions accumulate, use **Spend**, **Models**, **Tools**, **Efficiency**, and **Git** to review longer-term patterns. ## Product Tour ### Follow a session Run keeps usage, execution, tool, and budget evidence together on one focused session page. ### Understand spend Compare Today, 7-day, 30-day, This month, or a custom period across platforms, projects, runtimes, and sessions. Spend concentration, percentile session shapes, and a clickable cost-or-input versus active-time map expose which runs deserve inspection. ### Inspect tools and skills Review observed calls, output estimates, failures, repeats, catalog exposure, skill-pack activation, and bounded review candidates. ### Check token efficiency Use **Efficiency** to compare four signals over comparable, covered work: - **Output / $**: reported output tokens per covered dollar. Higher is better over time because more output is reaching the response for the spend. - **Reasoning ratio**: reported reasoning tokens as a share of output. Lower or stable is usually better for comparable work, while difficult work may need more reasoning. - **Context load**: processed input tokens per output token. Lower is better because less context is carried into each response. - **Output / execution**: output tokens per covered run. Higher generally means a less fragmented workflow. Each headline includes a daily trend, and partial coverage or unavailable evidence stays labelled beside the numbers. ### Git **Git** pairs successful local pushes with covered spend, so you can see code changed by project and day. It uses local `git` evidence only—never remote requests—and clearly marks partial or unavailable coverage. It is a mechanical signal, not a code-quality or productivity score. ### Configure budgets and agent access Manage monthly budgets, model pricing, language signals, native preferences, and local read-only connections for Codex and Claude. Software update checks and automatic installation are separate settings; both are on by default. The local MCP exposes seven read-only tools: | Tool | Use | | --- | --- | | `check` | Make a bounded decision about the caller-matched current run. | | `usage` | Review aggregate spend, model, tool, or change evidence. | | `capabilities` | Review optional user-installed skill-pack evidence. | | `sessions` | Select content-free session IDs using runtime, client, model, state, or time filters. | | `trace` | Read a standardized trace or sanitized runtime-native structure for one session. | | `stats` | Aggregate selected token, cost, timing, context, attempt, model-call, or tool metrics. | | `schema` | Discover fields, dimensions, units, limits, and availability semantics. | A comparison harness can call `sessions`, pass one returned ID to `trace`, and then call `stats` with dimensions such as `runtime`, `model`, `day`, or `session_id`. List responses expose `page.next_cursor`; continue by replaying the same query with that cursor. Metrics retain measured, estimated, inferred, and unavailable coverage, so missing evidence is not silently treated as zero. The `native_structure` trace view is not raw trace content. It keeps only allowlisted event types/subtypes, model and tool identities, statuses, relationships, timestamps, and numeric evidence. It does not expose raw trace content, prompts, responses, reasoning text, tool payloads, or trace paths. ### Check without opening the dashboard Use the macOS menu bar, Linux tray, or beta Windows extension to reach the current or pinned run. The native clients read a compact local payload and do not parse traces or read provider credentials directly. Token Meter checks for updates every 10 minutes and installs safe `main` updates automatically by default, so normal updates do not require opening the dashboard. If automatic installation is off, the native menu shows **New update available** instead. ## Evidence and Privacy Token Meter reads local runtime stores and binds its dashboard to `127.0.0.1`. It does not upload traces, prompts, responses, project paths, token counts, costs, or derived analytics. Do not expose the localhost dashboard publicly. Costs and selected token values can be estimates. Codex cost uses public API-equivalent rates, which can differ from subscription billing; Cursor usage includes local proxies where authoritative values are unavailable; Pi cost are estimates based on model API pricing. The optional MCP returns bounded derived evidence, not prompts, responses, reasoning, tool contents, credentials, settings, or trace paths. A result sent to an explicitly connected agent may be processed by that client's model provider under its own terms. See the [User guide](specs/USER_GUIDE.md) for the full evidence semantics and [Security policy](specs/SECURITY.md) for the canonical boundary. ## Documentation | Document | Use it for | | --- | --- | | [User guide](specs/USER_GUIDE.md) | Requirements, daily use, MCP, updates, uninstall, evidence semantics, and troubleshooting | | [Security](specs/SECURITY.md) | Privacy and security boundaries or vulnerability reporting | | [Architecture](specs/ARCHITECTURE.md) | Components, data flow, runtime adapters, and extension contracts | | [Contributing](specs/CONTRIBUTING.md) | Issues, pull requests, development, and validation | | [Product principles](specs/PRODUCT.md) and [visual design](specs/DESIGN.md) | Product and experience decisions | | [Specifications and plans](specs/) | Maintained feature designs and implementation plans | ## License MIT. See [LICENSE](LICENSE). ## Source & license This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [splunk](https://github.com/splunk) - **Source:** [splunk/token-meter](https://github.com/splunk/token-meter) - **License:** MIT - **Homepage:** https://splunk.github.io/token-meter/ Install and usage instructions live in the source repository linked above. ## Pricing - **Free** — Free ## Security capabilities Automated source analysis of v0.1.0 — what this tool can access: - **Network access:** no - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** no - **Dynamic code execution:** no *"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** — security scan: passed — Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/mcp-splunk-token-meter - Seller: https://agentstack.voostack.com/s/splunk - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.