Installation

$ npx @aaif/mcp-claims-triage@experimental

Try in Claude Desktop

claude_desktop_config.json

{
  "mcpServers": {
    "ai-claims-triage": {
      "command": "npx",
      "args": [
        "@aaif/mcp-claims-triage@experimental"
      ],
      "env": {}
    }
  }
}

Paste into ~/Library/Application Support/Claude/claude_desktop_config.json and restart Claude Desktop.

Tool

Description

triage_claim

Classify and route a new claim

get_triage_categories

List available triage categories and routing rules

explain_triage_decision

Return reasoning for a triage decision

Overview

The AI Claims Triage MCP server enables AI agents to classify, prioritise and route incoming insurance claims using configurable LLM-based triage logic. Reduces manual triage effort and accelerates claims handling SLAs.

How it works

Incoming claim data (FNOL details, loss description, coverage type) is passed to the MCP tool. The server applies a configurable classification pipeline — combining rule-based filters with LLM reasoning — and returns a triage decision with confidence score and recommended routing.

Use cases

FNOL triage — classify first notice of loss by complexity and coverage type
Fast-track routing — automatically route simple claims to straight-through processing
Fraud indicators — flag claims with anomalous patterns for manual review

Configuration

TRIAGE_LLM_MODEL=claude-sonnet-4-6 # or gpt-5-mini TRIAGE_RULES_PATH=./rules.yaml

Exposed MCP Tools

Tool	Description

`triage_claim`	Classify and route a new claim
`get_triage_categories`	List available triage categories and routing rules
`explain_triage_decision`	Return reasoning for a triage decision