AWS DR Cost Estimator MCP Server
Plan AWS disaster recovery budgets conversationally, comparing backup/restore, pilot light, warm standby, and active/active strategies using live Cost Explorer data.
README
AWS DR Cost Estimator MCP Server
This sample is provided for demonstration and educational purposes only and is not intended for production use without additional security review and testing.
Plan your AWS disaster recovery (DR) budget conversationally. Ask your MCP-compatible assistant (Kiro, Cline, Claude, etc.) what it would cost to run DR under backup and restore, pilot light, warm standby, or active/active for your workload. Compare all four strategies side by side in one response.
The server reads your current spend from the AWS Cost Explorer API, applies DR cost multipliers derived from real customer engagements, and returns a budgetary estimate in seconds — with per-row confidence levels so you know where the estimate is well-grounded and where it needs further validation.
For the full methodology, see docs/methodology.md.
Quick start
1. Install
Add the server with one click.
| Kiro | Cursor | VS Code |
|---|---|---|
Or paste this into your MCP client config manually:
{
"mcpServers": {
"dr-cost-estimator": {
"command": "uvx",
"args": ["--from", "git+https://github.com/aws-samples/sample-dr-cost-estimator-mcp.git", "dr-cost-estimator-mcp"],
"env": {
"FASTMCP_LOG_LEVEL": "WARNING",
"AWS_PROFILE": "your-aws-profile",
"AWS_REGION": "us-east-1"
}
}
}
}
uvx fetches and runs the server from this repo with no clone or build needed.
2. Set up IAM permissions
Attach a policy granting ce:GetCostAndUsage to the IAM identity behind your AWS_PROFILE:
{
"Version": "2012-10-17",
"Statement": [
{
"Sid": "DRCostEstimatorCostExplorerRead",
"Effect": "Allow",
"Action": "ce:GetCostAndUsage",
"Resource": "*",
"Condition": {
"StringEquals": {
"aws:RequestedRegion": "us-east-1"
}
}
}
]
}
Cost Explorer doesn't support resource-level ARNs for GetCostAndUsage, so Resource must be *. The condition restricts calls to the us-east-1 endpoint (the only endpoint Cost Explorer uses). For additional controls, consider SCPs or aws:PrincipalTag conditions.
3. Ask your assistant
"Compare all four DR strategies for my AWS account over the last 6 months."
That's it.
Example prompts
- "Compare all four DR strategies for my AWS account over the last 6 months and show me estimated costs."
- "Using my
prodAWS profile, analyze warm standby DR cost for the last 3 months." - "For AWS accounts
111122223333and444455556666, analyze pilot light DR cost over the last 3 months." - "Estimate DR costs for the workload tagged
Project=checkoutacross all my accounts." - "Compare DR strategies for resources tagged
app=paymentsorapp=ledgerin production (env=prod)." - "What would warm standby cost if we keep 30% warm capacity?"
Features
- Three tools:
list_dr_strategies,compare_dr_strategies,analyze_dr_strategy - Four DR strategies: backup/restore, pilot light, warm standby, active/active
- Live Cost Explorer input averaged over 1-24 complete months, optionally scoped by linked accounts or cost allocation tags
- Usage-type-aware multipliers for RDS, OpenSearch, and Elastic Load Balancing (compute vs. storage vs. backup get separate multipliers)
- Dollar-weighted confidence showing what fraction of spend has high-confidence vs. default-fallback multipliers
- Per-row rationale with confidence level and plain-language explanation
- Per-workload estimates via tags (
tag_filtersparameter) - Tunable warm-standby capacity (0.1-1.0) to match your RTO requirements
For detailed parameters, response shapes, and example calls, see docs/tool_reference.md.
Customizing multipliers
The DR cost multipliers live in JSON files under dr_cost_estimator_mcp/core/data/; edit them directly in your clone. For a readable rendering of every shipped multiplier with its rationale, see docs/multiplier_reference.md (generated from the data files). For guidance on tuning multipliers for your portfolio, see docs/methodology.md.
Troubleshooting
| Symptom | Fix |
|---|---|
| Server doesn't appear in tool list | Restart your MCP client after config changes. Set FASTMCP_LOG_LEVEL=DEBUG and check server logs. |
INVALID_INPUT error |
Check parameter ranges: strategy must be a valid id, top_n 1-500, months_back 1-24, account ids must be 12 digits. |
AWS_ERROR |
Verify AWS_PROFILE is set, the profile has the IAM policy above, and months_back covers a period with actual spend. |
| Empty estimates (no error) | All line items matched the skip list (tax, refunds, support, Savings Plans). This is expected when there's no estimable spend. |
| Numbers differ from Cost Explorer console | The tools use UnblendedCost at monthly granularity, excluding the current incomplete month. Console views using amortized cost or partial months will differ. |
Security
The server is read-only: ce:GetCostAndUsage is the only AWS API call. It cannot create, modify, or delete any resource. No data is written to disk; all output goes to stderr.
Documentation
- Methodology — multiplier tiers, assumptions, worked examples, customization guidance
- Tool Reference — parameters, response shapes, example calls
- Multiplier Reference — every shipped multiplier with rationale (generated)
License
This project is licensed under the MIT-0 License. See the LICENSE file for details.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
mcp-server-qdrant
这个仓库展示了如何为向量搜索引擎 Qdrant 创建一个 MCP (Managed Control Plane) 服务器的示例。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器