OpenClaw 的 Token 成本遥测工具，可从任何 Claude 或支持 MCP 的代理进行查询。 提供按代理 + 按提供商的归因、成本激增异常检测、模型路由建议以及 30 天预测。是 silentwatch-mcp 和 openclaw-health-mcp 的配套工具。
生产环境中的 AI 部署会在多个提供商、数十个代理和数百次技能调用中产生 Token 成本。账单在月底才会送达——到那时，想要回答“钱花哪儿了？”这个问题，就得从 10 万行日志中去挖掘。此 MCP 服务器实时呈现这些数据，并允许在 Claude 内部用自然语言进行查询：
> claude: where did our LLM spend go this week?
[MCP tools: cost_overview + top_cost_drivers]
Total spend last 7 days: $42.18
By provider:
Anthropic $30.40 (72%) — claude-sonnet-4 dominates
OpenAI $9.20 (22%) — chat-bot agent
Gemini $1.86 (4%) — cron-summarizer (cheap-route working)
Ollama $0.00 (local, free)
Top 3 cost drivers:
data-extraction-agent $28.50 (68%)
chat-bot $9.20 (22%)
cron-summarizer $1.86 (4%)
1 anomaly flagged — see find_cost_anomalies for details.> claude: any cheaper-routing opportunities?
[MCP tool: model_routing_recommendations]
Recommendation: data-extraction-agent currently runs claude-sonnet-4
with avg 400-token completions — extraction-style work that
gemini-2.5-flash usually handles at ~95% quality for ~5% the cost.
Estimated savings: $27.10/30d if migrated. Test on a 10% slice first.Related MCP server: nikhilnt
openclaw-cost-tracker-mcp现有工具（提供商账单仪表板、通用 FinOps 工具、自定义脚本）无法做到以下三点：
按代理归因，而不仅仅是按提供商汇总。 提供商仪表板显示“在 Anthropic 上花费了 $X”——它们无法告诉你六个代理中哪一个导致了 78% 的支出。成本追踪器读取 OpenClaw 的每请求成本日志 JSONL，并在保持 agent_id 不变的情况下进行聚合。
按代理的成本激增异常检测。 在聊天中粘贴一次 12 万 Token 的内容，其成本可能超过一周的正常流量。默认的“代理中位数 3 倍”阈值会在这些异常出现在月底账单之前就将其标记出来。
基于实际使用情况的路由建议。 通用的“使用更便宜的模型”建议毫无用处。此服务器可以识别出那些通过其“用量 + 完成长度”模式表明使用更便宜的提供商即可获得相同结果的特定代理，并提供具体的 30 天节省估算。
专为在生产环境中运行 OpenClaw 且有实际支出压力的 生产 AI 运维人员 而设计。
工具 | 返回内容 |
| 指定时间窗口内的总支出 + 按提供商分类 + 顶级代理 + 顶级模型 + 异常计数 |
| 按代理细分，包含平均单次请求成本 + 总占比 |
| 按提供商细分，包含 Token 计数 |
| 被标记为超过其代理中位数成本 3 倍以上的请求 |
| 支出最高的前 N 个代理 + 模型，无其他干扰信息 |
| 具体的更便宜模型建议，附带 30 天节省估算 |
| 预测 30 天总支出 + 按提供商分类，并附带置信度说明 |
资源：
cost://overview — 7 天快照
cost://forecast — 30 天预测
cost://anomalies — 最近标记的异常
提示词：
diagnose-cost-spike — 追踪最近的成本激增至根本原因 + 纠正措施
weekly-cost-digest — 200 字的每周成本摘要
pip install openclaw-cost-tracker-mcp添加到 ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) 或 %APPDATA%\Claude\claude_desktop_config.json (Windows)：
{
"mcpServers": {
"openclaw-cost": {
"command": "python",
"args": ["-m", "openclaw_cost_tracker_mcp"],
"env": {
"OPENCLAW_COST_BACKEND": "mock"
}
}
}
}后端 | 状态 | 描述 |
| ✅ v1.0 | 包含故意设置的异常 + 路由机会的示例数据，用于协议验证 |
| ✅ v1.0 | 解析 OpenClaw 原生的成本日志 JSONL 文件（默认 |
| ⏳ v1.1 | 直接读取 Anthropic + OpenAI 账单 API（无需日志垫片） |
每一行都是一条 JSON 记录：
{"request_id":"req-abc123","timestamp":"2026-05-04T12:34:56Z","provider":"anthropic","model":"claude-sonnet-4","agent_id":"data-extraction-agent","skill_id":"extract-structured-data","prompt_tokens":8500,"completion_tokens":600,"total_tokens":9100,"cost_usd":0.0345,"duration_ms":4823}如果您的 OpenClaw 部署不输出此格式，请使用小型日志垫片包装您的提供商调用——示例垫片位于 examples/ 中。
版本 | 范围 | 状态 |
v1.0 | mock + openclaw-jsonl 后端，7 个工具 / 3 个资源 / 2 个提示词，异常检测 + 路由 + 预测，GitHub Actions CI 矩阵，PyPI 可信发布 | ✅ |
v1.1 |
| ⏳ |
v1.2 | 后端联邦；预算警报 + 阈值突破检测 | ⏳ |
v1.x | 按渠道成本归因；预算警报的 Webhook 发送器 | ⏳ |
如果您的 AI 部署不使用 OpenClaw 的成本日志格式——例如使用不同的代理工具、自定义日志记录、AWS Bedrock 计量、供应商账单 API——而您想要同样的归因 + 异常 + 路由可见性，这属于 自定义 MCP 构建 服务。
层级 | 范围 | 投资 | 时间线 |
简单 | 为您现有的成本数据源提供单个后端适配器 | $8,000–$10,000 | 1–2 周 |
标准 | 自定义后端 + 自定义异常规则 + 与您的警报系统集成 | $15,000–$20,000 | 2–4 周 |
复杂 | 多后端联邦 + 预算执行 + 自定义路由逻辑 | $25,000–$35,000 | 4–8 周 |
如何联系：
发送电子邮件至 temur@pixelette.tech，主题为 Custom MCP Build inquiry
内容包括：一段关于您技术栈的描述 + 您正在考虑的层级
我们将在 2 个工作日内回复，并提供 30 分钟的初步沟通时段
此服务器是 生产 AI 基础设施 MCP 套件 的一部分——配套工具包括 silentwatch-mcp（Cron 静默故障检测）和 openclaw-health-mcp（部署健康状况）。安装全部三个以获得全面的运营可见性。
如果您正在运行生产级 AI，并希望外部从业者评估就绪情况、发现现有的故障模式（成本超支是最常见的模式之一）并编写纠正行动计划：
层级 | 范围 | 投资 | 时间线 |
审计精简版 | 单个系统，前 5 大发现，书面报告 | $1,500 | 1 周 |
审计标准版 | 全面审计，涵盖所有 14 种模式，5C 发现，90 天后续跟进 | $3,000 | 2–3 周 |
审计 + 工作坊 | 标准审计 + 2 天团队工作坊 + 包含首次月度审计 | $7,500 | 3–4 周 |
联系方式同上：temur@pixelette.tech，主题为 AI audit inquiry。
欢迎提交 PR。后端是可插拔的——请参阅 src/openclaw_cost_tracker_mcp/backends/ 获取契约。
添加新后端：
在 backends/<your_backend>.py 中继承 CostBackend
实现 get_entries()
在 backends/__init__.py 中注册
在 tests/test_backend_<your_backend>.py 中添加测试
Bug 报告 + 功能请求：请提交 GitHub Issue。
MIT — 请参阅 LICENSE。
silentwatch-mcp — Cron 静默故障检测
openclaw-health-mcp — 部署健康状况
AI Production Discipline Framework — Notion 模板，$29 — 这些 MCP 工具所实现的方法论（成本超支是目录中的 P3.x 模式）
SPEC.md — 完整的服务器设计
由 Temur Khan 构建 — 生产 AI 系统独立从业者。 联系方式：temur@pixelette.tech
Top-level cost summary for a window — total spend, request count, by-provider breakdown, top agents, top models, anomaly count. Use this first for a single-pane view.
| Name | Required | Description | Default |
|---|---|---|---|
| window_hours | No | Lookback window in hours (default 168 = 7 days, max 720 = 30 days) | |
| top_n | No | How many top agents + models to include (default 5) |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations provided, so description carries the burden. Description implies read-only behavior (cost summary), but does not explicitly state non-destructiveness, rate limits, or other traits.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Two sentences with zero waste. Front-loaded with output items. Every sentence earns its place.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
No output schema, but description lists all expected return components (total spend, request count, breakdowns, top agents/models, anomalies). Adequate for a top-level summary tool.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100% (both parameters described). Description adds no additional meaning beyond what the schema provides, so baseline 3 is appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
Description clearly states verb (summary) and specific resource (top-level cost for a window). It lists the exact components returned and distinguishes from siblings via 'use this first for a single-pane view'.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Explicitly says 'Use this first for a single-pane view', implying it's the entry point. No explicit when-not or alternatives, but the sibling list provides context for other tools.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Per-agent cost breakdown — total spend + request count + avg cost-per-request + primary provider/model + share of total spend, sorted by cost descending.
| Name | Required | Description | Default |
|---|---|---|---|
| window_hours | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are provided, so the description carries the full burden. It does not disclose whether the tool is read-only, destructive, or has auth/rate limits. Since it likely just queries data, a note about idempotency would be beneficial.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single sentence that efficiently lists all output fields and the sort order. No unnecessary words.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
The description explains the output well but omits the single input parameter entirely. Given no output schema, this is a notable gap for input comprehension.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The input schema has one optional parameter (window_hours) with 0% schema description coverage. The description does not mention this parameter at all, failing to explain its meaning or effect on results. The description adds no value beyond the schema.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states it provides a per-agent cost breakdown with specific metrics (total spend, request count, avg cost-per-request, primary provider/model, share of total spend, sorted by cost descending). This distinguishes it from sibling tools like cost_overview (likely overall) or costs_by_provider.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The usage is implied from the purpose (use when you need per-agent costs), but there is no explicit guidance on when to use this tool versus alternatives, no prerequisites, and no mention of when not to use it.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Per-provider cost breakdown — total spend + request count + token counts + share of total, sorted by cost descending.
| Name | Required | Description | Default |
|---|---|---|---|
| window_hours | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations provided, the description is responsible for behavioral disclosure. It mentions output metrics and sorting but omits key details such as that it is a read-only operation, how the time window (window_hours) affects results, or any limits/pagination.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single concise sentence that front-loads the key information (what the tool returns and how it is sorted), with no superfluous content.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the tool has one optional parameter, no output schema, and several siblings, the description fails to explain the parameter or provide usage guidance, making it incomplete despite adequately covering purpose and basic output.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The input schema has one parameter (window_hours) with 0% description coverage in the schema. The description does not mention window_hours at all, leaving the agent unaware of how to specify the time range for the breakdown.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states it provides a per-provider cost breakdown with specific metrics (total spend, request count, token counts, share of total) sorted by cost descending, which distinguishes it from siblings like costs_by_agent and cost_overview.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description implies the tool is for per-provider breakdowns but does not explicitly state when to use it over alternatives like costs_by_agent or cost_overview, nor does it provide exclusions or context for when it is appropriate.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Requests whose cost is threshold_multiplierx or more above their agent's median request cost. Default 3.0x. Surfaces giant-context-paste accidents, runaway loops, model regressions, etc.
| Name | Required | Description | Default |
|---|---|---|---|
| window_hours | No | ||
| threshold_multiplier | No | Multiplier above agent median to flag (default 3.0, min 1.5) |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are provided, so the description carries full burden. It explains the tool flags requests exceeding a threshold, but does not disclose if it modifies data, requires permissions, or the format of returned anomalies. This is adequate but leaves gaps.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Two sentences suffice to convey purpose, examples, and default values. Information is front-loaded and no unnecessary words are present.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given no output schema, the description could elaborate on the output format. However, the context of sibling cost tools and the clear examples provide sufficient completeness for a diagnostic tool.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 50% with only threshold_multiplier having a description. The tool description adds meaning by explaining the purpose of threshold_multiplier (default 3.0x, min 1.5) and implies window_hours as a time window. This provides useful context beyond the schema.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description uses a specific verb 'find' and resource 'cost anomalies', clearly distinguishes from sibling tools like cost_overview or top_cost_drivers, and provides concrete use cases (giant-context-paste accidents, runaway loops).
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description implies usage for detecting anomalous cost spikes with examples, but does not explicitly state when not to use it or compare to alternatives. However, the context of sibling tools makes its purpose distinct.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Projects 30-day total spend from the observed window's run rate. Returns per-provider 30d projection + confidence note. Re-run with longer window for stability.
| Name | Required | Description | Default |
|---|---|---|---|
| window_hours | No | Hours of observed data to project from (default 168 = 7 days) |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are provided, so the description carries full burden. It discloses that the tool returns a per-provider projection and confidence note and suggests re-running for stability, but does not explicitly state whether the operation is read-only or discuss any side effects or permissions needed.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is three short, front-loaded sentences with no filler. Each sentence adds value: purpose, return structure, and usage advice. It is efficient and well-structured.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a tool with one optional parameter and no output schema, the description adequately covers what it does and returns (per-provider projection, confidence note). It is mostly complete, though it could specify the format of the projection or confidence level more precisely.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The input schema already provides 100% coverage (window_hours with description and default). The description adds no new parameter meaning beyond the schema, only context about stability. Baseline 3 is appropriate as the description does not detract but adds minimal extra value.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool projects 30-day total spend from observed run rate, with a specific verb ('projects') and resource ('30-day total spend'). It distinguishes from sibling tools like cost_overview or costs_by_provider by focusing on forecasting rather than current breakdowns.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description advises re-running with a longer window for stability, implying when to adjust the parameter, but it does not explicitly state when to use this tool versus alternatives like find_cost_anomalies or top_cost_drivers. No when-not or comparator guidance is provided.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Suggestions to route some volume to cheaper providers/models — e.g., claude-sonnet-4 → gemini-2.5-flash for extraction-style work, gpt-4o → gpt-4o-mini for short chats. Each rec includes 30d estimated savings.
| Name | Required | Description | Default |
|---|---|---|---|
| window_hours | No | Window over which to find candidates (default 720 = 30 days) |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are provided, so the description carries the full burden. It lacks details on side effects (e.g., does it modify anything?), permissions, limitations, or how suggestions are generated. The description only states output (suggestions with savings) but not behavior beyond that.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is two sentences, front-loaded with the main purpose, and uses concrete examples to clarify. Every sentence adds value without redundancy.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
The description is adequate for a recommendation tool but lacks details on output format, number of suggestions, or prerequisites (e.g., required cost data). Without output schema, more behavioral context would be helpful for an agent to use it correctly.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The single parameter 'window_hours' is fully described in the schema (coverage 100%). The description adds no additional meaning beyond what the schema provides, so baseline score of 3 applies.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states it provides suggestions to route volume to cheaper providers/models, with specific examples and mention of 30d estimated savings. It is easily distinguishable from sibling tools which focus on cost analysis rather than recommendations.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description does not explicitly state when to use this tool versus alternatives. While the sibling tool names give implicit context, there is no guidance on when to recommend routing vs. other cost actions, or exclusions for certain scenarios.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Highest-spend agents + models in the window — flat list, no per-provider or anomaly noise. Useful for 'where's our money going' digest format.
| Name | Required | Description | Default |
|---|---|---|---|
| window_hours | No | ||
| top_n | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations provided, so description carries full burden. It discloses that results are a flat list without per-provider or anomaly noise, which is sufficient for a simple query tool. Lacks mention of data freshness or side effects, but none expected.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
One concise sentence plus a usage suggestion; no filler words. Front-loaded with key information.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given low complexity (2 optional params, no output schema), description adequately covers purpose, format, and use case. Could add details on output fields or ordering, but not critical for a digest.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 0%, so description must compensate. It does not explicitly explain window_hours or top_n parameters, only implying a time window and top N via context. Defaults are in schema but not described.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
Description explicitly states it identifies 'highest-spend agents + models' and contrasts with siblings by saying 'flat list, no per-provider or anomaly noise,' making purpose and differentiation clear.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Explicitly suggests use case: 'where's our money going' digest format, and implicitly advises against using when per-provider breakdowns or anomalies are needed, referencing sibling tools.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
v1.0.0cost_overviewcosts_by_agentcosts_by_providerfind_cost_anomaliesforecast_monthly_costmodel_routing_recommendationstop_cost_driversScored across 7 tools
Each tool targets a distinct cost analysis aspect: overview, agent/provider breakdown, anomalies, forecasting, recommendations, and top drivers. No overlap in functionality.
All tool names use consistent lowercase underscore pattern, clearly indicating their purpose (e.g., cost_overview, find_cost_anomalies). Pattern is predictable and readable.
Seven tools is ideal for a focused cost tracker; each tool covers a specific need without redundancy or bloat.
Covers core cost analysis needs (overview, breakdowns, anomalies, forecasting, optimization). Minor gaps like filtering or export are acceptable given the scope.
Cost per run, tool, model and day for AI agents, from one call or an OpenTelemetry export.
Analytics for MCP servers. Query your tool calls, first-call success, retries and schema cost.
Agent Cost Allocator MCP — multi-tenant LLM cost attribution for chargeback billing. Companion to
Costory is the FinOps agent your team builds on. Claude plus a single provider's MCP works for one person and one cloud. It breaks when you add more providers, more people, and more tools to maintain. - Multiple providers: AWS, GCP, Azure, LLM APIs and data clouds normalized and allocated to teams and features in one model, including allocation from telemetry (Kubernetes, Datadog, dbt). - Multiple users: one hosted setup with OAuth, per-user access and enterprise SSO, instead of local credentials and prompts per person. - Maintained for you: tools, skills and connectors kept up to date by Costory, so reorgs and new services don't break your agent. Your agent finds savings, explains every change in spend and issues recommendations that your coding agent can implement. It works in Claude, Cursor and any MCP client, plus Slack. Customers include Alan, Deezer, Delivery Hero, Voodoo and Pennylane. Pricing: Starter from $300/month, flat: unlimited users and data sources, and a fair use that 80% of customers never outgrow. Nothing is ever cut off, and agents connected through the MCP use no AI credits. 30-day free trial.
