CRIC Wuye AI MCP Server for Property Management

Overview

CRIC Wuye AI MCP Server is an implementation of a Model Context Protocol (MCP) server tailored for property management workflows. It exposes a lightweight runtime that lets an LLM (or other reasoning agent) discover and invoke domain-specific tools (APIs) such as tenant lookup, maintenance ticketing, building analytics, and knowledge-base queries. By providing standardized tool manifests and runtime execution, the server helps integrate LLM-assisted agents into existing property management systems safely and predictably.

This server is useful when you want an LLM to perform actionable tasks rather than only generate text: it centralizes tool registration, exposes a machine-readable API surface for the model, and handles invocation, authentication, and request/response logging. Developers can run the MCP server locally or in production, register custom tools that wrap internal services, and connect an LLM that supports MCP-style tool calls to automate tenant support, maintenance routing, and administrative workflows.

Features

• MCP-compliant tool registry: serves machine-readable manifests so agents can discover available tools.
• Tool execution bridge: executes registered tool handlers and returns structured responses to the agent.
• Extensible plugin model: add new tools by following a simple manifest + handler convention.
• API authentication and basic logging: supports environment-based API keys and request tracing for audit.
• Docker and local development support: run quickly with Docker or node-based local start.
• Example tools for property management: tenant lookup, maintenance ticket creation, building stats, document retrieval.

Installation / Configuration

Clone the repository and install dependencies (Node.js 18+ recommended):

git clone https://github.com/wuye-ai/mcp-server-wuye-ai.git
cd mcp-server-wuye-ai
npm install

Create a .env file for basic configuration:

# .env
PORT=3000
API_KEY=replace-with-a-secret
LOG_LEVEL=info

Start locally:

npm run build
npm start
# or for development
npm run dev

Run with Docker:

# build docker image
docker build -t wuye-mcp-server .

# run container
docker run -e PORT=3000 -e API_KEY=replace-with-a-secret -p 3000:3000 wuye-mcp-server

Configuration notes:

• PORT: HTTP port the MCP server listens on.
• API_KEY: a shared secret used for simple authorization of tool registration and invocation; replace with your vault-managed secret in production.
• LOG_LEVEL: controls verbosity (e.g., debug, info, warn, error).

Available Resources

The server exposes a small set of HTTP endpoints useful when connecting an LLM or developer tooling. Typical endpoints:

Example: fetch tools with curl

curl -H "Authorization: Bearer replace-with-a-secret" http://localhost:3000/tools

Tool manifests are stored in the repository under a configurable directory (e.g., ./tools). Each manifest includes metadata, input schema, and a description to help an LLM decide when to call it. Handlers map those manifests to real business logic (HTTP calls, DB queries, message queues).

Use Cases

• Tenant support automation

  • Scenario: an LLM assistant takes tenant natural-language requests (“My hot water isn’t working”) and, using the MCP server, calls the createMaintenanceTicket tool with structured fields (tenantId, unit, issueType, priority). The server invokes your ticketing API and returns the ticket number to the assistant, which replies to the tenant.

• Maintenance triage and routing

  • Scenario: incoming sensor alerts are summarized by an LLM; the assistant consults /tools to find diagnoseSensorEvent, runs diagnostics through the MCP server, and if necessary, schedules a vendor via scheduleServiceVisit.

• Document retrieval and compliance

  • Scenario: an LLM answers compliance questions by invoking fetchLeaseDocument or searchKB tools exposed through the MCP server. The tools return structured excerpts or document IDs rather than raw untrusted text.

• Daily operations and reporting

  • Scenario: an operations agent asks for “open tickets for Building A this week” and calls queryTickets via the MCP server which returns a summarized JSON that the assistant converts into human-readable status updates.

Getting Started Tips

1. Start with simple read-only tools (search, read) to test the discovery and invocation flow.
2. Use input schemas in tool manifests to limit and validate tool inputs coming from the LLM.
3. Instrument logging for all tool invocations for auditing and quality control.
4. Gate destructive tools (e.g., cancelLease, deleteRecord) behind stronger authentication/approval flows.

Repository: https://github.com/wuye-ai/mcp-server-wuye-ai — consult the repo for example manifests, handlers, and Docker setup.