stateless-mcp-incident-lab-typescript-raw
A raw TypeScript MCP server simulating incident lifecycles with tools, resources, prompts, and signed multi-round remediation approval. Enables CLI-based interaction, protocol negotiation, and health/security testing without external MCP SDK dependencies.
README
Stateless MCP Incident Lab — Raw TypeScript
Independent raw HTTP/JSON-RPC implementation of the Stateless MCP Incident Lab for Node.js 24. It implements MCP wire shapes directly and intentionally does not depend on the official MCP SDK.
Capabilities
- Stateless per-request protocol-version and capability negotiation
- Streamable HTTP JSON responses and SSE diagnostic progress
- Tool, resource, resource-template, and prompt catalogs
- Synthetic incident lifecycle and signed multi-round remediation approval
- Public/private response caching with stale-on-refresh-error behavior
- CLI workflows, structured telemetry, health readiness, and security bounds
All remediation effects are simulated.
Quick start
npm ci
npm run build
npm start
The server binds to 127.0.0.1:3101 by default and exposes POST /raw/mcp plus GET /raw/healthz. Override the bind with HOST and PORT. Production replicas must share an MCP_REQUEST_STATE_SECRET of at least 32 bytes and set DYNAMODB_TABLE (plus optional DYNAMODB_ENDPOINT for DynamoDB Local); development uses process-local ephemeral key and effect storage. In another shell:
node dist/src/main.js --version
node dist/src/main.js discover http://127.0.0.1:3101/raw/mcp
CLI
The incident-mcp binary supports these command families:
incident-mcp discover <url>
incident-mcp tools list|inspect|call <url> [...]
incident-mcp resources list|templates|read <url> [...]
incident-mcp prompts list|get <url> [...]
incident-mcp demo <url> --approve|--decline|--cancel
Global diagnostic options include --wire with sensitive-value redaction and --no-cache.
Docker
Build and run the minimal runtime image:
docker build --target runtime -t incident-mcp-raw .
docker run --rm -p 127.0.0.1:3101:3101 \
-e HOST=0.0.0.0 \
-e MCP_REQUEST_STATE_SECRET="$(openssl rand -hex 32)" \
-e EFFECT_STORE=memory \
incident-mcp-raw
# Or inspect the image version without starting the service:
docker run --rm incident-mcp-raw --version
The test stage requires the external behavioral repository as a named build context:
docker build --target test \
--build-context conformance=../stateless-mcp-incident-lab-conformance \
-t incident-mcp-raw-test .
Runtime behavior
The public HTTP/CLI path uses the same application modules exercised by conformance. Incident lifecycle records and conditional remediation claims share the configured DynamoDB table in production; development uses in-memory stores. The CLI applies MCP cache hints in-process, supports explicit bypass, and serves stale entries only after refresh failure with a warning. Diagnostic SSE work is paced and request-scoped, stops on disconnect or deadline, and returns the actual diagnostic result. Live JSON requests enforce a five-second deadline and emit structured, trace-correlated telemetry with bearer-bearing resource names redacted.
Development and verification
npm run format
npm run lint
npm run typecheck
npm run test:coverage
npm run test:mutation
npm run build
npm run test:conformance
npm audit --audit-level=high
The behavioral suite is resolved from CONFORMANCE_PATH, defaulting to the sibling ../stateless-mcp-incident-lab-conformance/conformance directory.
Coverage is enforced at 95% lines, 90% statements, 80% branches, and 95% functions. The remaining lines are defensive malformed-internal-input paths and the direct-process bootstrap. Mutation testing covers incident lifecycle, MRTR, cache, property laws, header/request-state security, and schema validation, with an enforced score of at least 80%.
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。
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。
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。
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。
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。
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。
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
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。