niro-mcp-server

An MCP (Model Context Protocol) server that exposes AU-SG Retail BI domain intelligence as tools for Kiro and other MCP-compatible AI clients.

What It Does

NIRO MCP Server acts as a virtual BIE — it knows your metrics, dashboards, data lineage, and team conventions. Install it once and any MCP client (Kiro, Claude Desktop, etc.) can:

• Look up metrics — definitions, formulas, source tables, ETL jobs, aliases
• Classify intent — determine if a request is a CHANGE, STATUS, or DATA query
• Triage requests — find duplicates, suggest assignees, recommend actions
• Generate proposals — structured solution proposals with draft SQL and effort estimates
• Trace lineage — upstream/downstream data flow from any table, ETL, or dashboard
• Assess impact — what breaks if a source table or ETL changes
• Check freshness — expected refresh schedules and acceptable lag
• Detect deprecation — find deprecated sources and their replacements
• Search domain knowledge — full-text search across all BI knowledge
• Create SIM tickets — file new requests in the AU-Retail-BI resolver group
• Synthesize status — unified view of SIM tickets, ETL health, and sprint progress

Installation

pip install niro-mcp-server

Or run without installing (requires uv):

uvx niro-mcp-server

Configuration for Kiro

Add to your .kiro/settings/mcp.json:

{
  "mcpServers": {
    "niro": {
      "command": "niro-mcp-server",
      "env": {"FASTMCP_LOG_LEVEL": "ERROR"},
      "autoApprove": [
        "niro_lookup_metric",
        "niro_classify_intent",
        "niro_find_similar",
        "niro_trace_lineage",
        "niro_impact_analysis",
        "niro_check_freshness",
        "niro_check_deprecation",
        "niro_domain_search",
        "niro_synthesize_status"
      ]
    }
  }
}

Or use uvx for zero-install:

{
  "mcpServers": {
    "niro": {
      "command": "uvx",
      "args": ["niro-mcp-server"],
      "env": {"FASTMCP_LOG_LEVEL": "ERROR"}
    }
  }
}

Prerequisites

• Python 3.10+
• Midway authentication (optional) — needed for SIM/Taskei operations. Run mwinit to authenticate. All read-only tools work without it.
• AWS credentials (optional) — needed for LLM-powered proposals via Amazon Bedrock. Without credentials, proposals use template-based generation.

Available Tools

License

MIT