Andiora Developer MCP
Enables developers to read project context, create documentation drafts, update reports, and retrieve tenant-scoped metrics through the Andiora platform APIs.
README
Andiora Developer MCP
Design documentation and implementation guidance for the Andiora Tech Model Context Protocol (MCP) server.
The MCP server will provide a governed interface for developer assistants to read project context, create documentation drafts, update operational reports, and retrieve tenant-scoped metrics through the Andiora platform APIs.
Documentation
Engineering principles
- Reuse the Andiora API and authorization model; never connect the MCP server directly to PostgreSQL.
- Enforce tenant and project boundaries on every tool invocation.
- Keep read operations separate from write operations and require explicit confirmation for consequential actions.
- Record tool calls and mutations in an auditable event stream.
- Validate all inputs with strict schemas, apply idempotency to writes, and avoid returning secrets.
- Start with a local stdio transport for developers; add authenticated HTTPS transport only after the security controls are complete.
Repository status
Local development
npm install
npm run build
npm start
No .env file is required for the standard Andiora dev environment. On the
first run, the MCP starts a loopback OAuth callback on 127.0.0.1:4318, opens
the Cognito managed login in the browser, validates the Authorization Code +
PKCE response, and exchanges the Cognito identity for a short-lived Andiora
access token. The token is kept in memory only. .env is reserved for private
stages or advanced overrides.
# Optional private-stage overrides only:
ANDIORA_API_URL=https://your-api.example.com
MCP_OIDC_ISSUER=https://cognito-idp.us-east-1.amazonaws.com/<pool-id>
MCP_OIDC_CLIENT_ID=<public-client-id>
The first runtime phase intentionally exposes only read tools over stdio. It uses the existing Andiora REST API, applies a configured tenant boundary, validates inputs, redacts sensitive logs, and sends idempotency support for future write tools.
Use list_projects first to discover project IDs visible to the authenticated employee, then pass a selected projectId to get_project_roadmap. Project visibility and permissions remain enforced by the Andiora API.
The read-only catalog currently includes get_organization_status, list_projects, get_project_roadmap, list_documents, list_deliverables, list_tickets, get_analytics_summary, get_health_diagnostics, list_monitors, and list_team_members. Billing and Vault are intentionally excluded from the general developer MCP because they expose sensitive financial or infrastructure context.
The create_markdown_document tool uploads UTF-8 Markdown to the Andiora Document Hub as a DRAFT. It enforces a safe filename, a 2 MB content limit, tenant ownership, short-lived presigned S3 upload URLs, and idempotency keys. Publication remains a human-controlled platform workflow.
The workflow tools are create_markdown_version, submit_document_for_review, approve_document, publish_document, and create_report_draft. Versions are linked to their parent document, reports use the same sanitized Markdown pipeline, transitions are enforced by the platform, and the MCP applies a local limit of 60 API requests per minute. Structured audit logs are emitted without tokens or secrets.
Client configuration
For Cursor, Claude Desktop, or another stdio-capable client, configure the compiled entry point:
{
"mcpServers": {
"andiora-dev": {
"command": "node",
"args": ["/absolute/path/to/andiora-dev-mcp/dist/index.js"]
}
}
}
The MCP OAuth flow is intentionally local stdio plus a loopback callback. It
does not expose an HTTP MCP server or persist tokens. Cognito provisioning and
the password-migration trigger are defined in the Andiora platform repository;
see its docs/MCP_COGNITO_AUTH.md.
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。