Customer Service MCP Server

Customer service platform for AI agents — conversations, customer intelligence, knowledge base, response suggestions, churn prediction, routing, and satisfaction tracking. 20 tools with realistic demo data.

Architecture

MCP Customer Service Architecture

Tools (20)

Conversations (5)

| Tool | Purpose | Risk | |------|---------|------| | list_conversations | Filter by status, priority, agent | read_only | | get_conversation | Full thread with messages | read_only | | start_conversation | Start new conversation | internal_write | | reply_conversation | Public reply to customer | external_write | | add_internal_note | Internal note (not visible) | internal_write |

Customer Intelligence (4)

| Tool | Purpose | Risk | |------|---------|------| | get_customer_profile | Plan, LTV, health, history | read_only | | get_customer_health | Health score + contributing factors | read_only | | get_interaction_history | All past conversations | read_only | | assess_churn_risk | Churn prediction with signals | read_only |

Resolution (4)

| Tool | Purpose | Risk | |------|---------|------| | search_knowledge_base | Find relevant help articles | read_only | | suggest_response | AI-suggested reply based on context | read_only | | get_canned_responses | Approved response templates | read_only | | resolve_conversation | Mark resolved + trigger CSAT | internal_write |

Routing & Escalation (3)

| Tool | Purpose | Risk | |------|---------|------| | assign_agent | Assign to agent or team | internal_write | | escalate | Escalate with reason | internal_write | | get_queue_status | Queue depth, wait times | read_only |

Metrics & Management (4)

| Tool | Purpose | Risk | |------|---------|------| | get_satisfaction_scores | CSAT, NPS, effort scores | read_only | | get_service_metrics | Response time, FCR, volume | read_only | | list_agents | Agent availability | read_only | | merge_conversations | Merge duplicates | destructive |

Installation

cargo install mcp-customer-service

Configuration

The MCP server is a thin API client. It requires a backend that implements the Customer Service API Spec.

| Env Var | Required | Description | |---------|:---:|-------------| | CUSTOMER_SERVICE_API_URL | ✅ | Base URL of your backend (e.g. http://localhost:8080/api/v1) | | CUSTOMER_SERVICE_API_KEY | ❌ | Bearer token for auth |

Supported Backends

| Backend | How | |---------|-----| | Your own API | Implement the API spec in any language | | Zendesk | Build an adapter that maps Zendesk API → this spec | | Freshdesk | Build an adapter that maps Freshdesk API → this spec | | Intercom | Build an adapter that maps Intercom API → this spec |

The MCP server doesn't store any data — it's a pure passthrough to your backend.

Client Configuration

{
  "mcpServers": {
    "customer-service": {
      "command": "mcp-customer-service",
      "args": [],
      "env": {
        "CUSTOMER_SERVICE_API_URL": "http://localhost:8080/api/v1",
        "CUSTOMER_SERVICE_API_KEY": "your-api-key"
      }
    }
  }
}

AI Agent Workflow

1. Triage    → list_conversations(status="open") → classify priority
2. Understand → get_customer_profile + get_customer_health → context
3. Resolve   → search_knowledge_base + suggest_response → draft reply
4. Respond   → reply_conversation (or escalate if complex)
5. Close     → resolve_conversation → CSAT survey sent

Usage Examples

Handle an angry customer

"There's an urgent billing complaint"
→ list_conversations(status="open", priority="urgent")
→ get_conversation(id="conv-3")
→ get_customer_profile(id="cust-4") — Free plan, health 30, declining
→ assess_churn_risk(id="cust-4") — HIGH risk, 5 signals
→ suggest_response(id="conv-3") — "I understand your concern about the charge..."
→ reply_conversation(id="conv-3", body="...")

Proactive churn prevention

"Which customers are at risk of churning?"
→ assess_churn_risk for each customer
→ start_conversation for high-risk ones with proactive outreach

Demo Data Highlights

| Customer | Plan | Health | Sentiment | Churn Risk | |----------|------|:------:|-----------|:----------:| | Sarah Mitchell | Enterprise | 85 | Stable | Low | | James Park | Pro | 45 | Declining | High | | Maria Garcia | Enterprise | 92 | Improving | Low | | Tom Wilson | Free | 30 | Declining | High | | Lisa Chen | Pro | 72 | Stable | Medium |

MCP Server Manifest

server_id = "mcp_customer_service"
display_name = "Customer Service"
version = "1.0.0"
domain = "customer-service"
risk_level = "medium"
writes_allowed = "gated"

License

Apache-2.0

Part of the ADK-Rust Enterprise MCP server ecosystem.

Built with ❤️ by Zavora AI

MCP Servers