Customer Support Automation Agent

Connect your helpdesk and knowledge base via MCP

Create MCP tools that give the agent controlled access to your support systems.

Tools to expose via MCP:
- search_knowledge_base(query) → returns relevant help articles
- get_customer_account(email_or_id) → account status, history
- get_ticket_history(customer_id) → previous tickets and resolutions
- get_return_policy() → current policy text
- create_ticket(customer_id, subject, description) → escalate with context
- check_refund_eligibility(order_id) → policy-based eligibility check

Define escalation logic in the MCP layer

Your MCP server can apply business rules before returning data — for example, automatically marking a ticket as 'escalation needed' if the issue matches certain criteria, giving the agent clear signals on when to hand over.

Load your policies and help docs

In addition to live API tools, upload your return policy, warranty terms, and FAQ content to the agent's knowledge base. The agent searches both live data (via MCP) and static knowledge when answering.

Create the Support Agent

Write a system prompt that defines the agent's tone, capabilities, and escalation rules.

System Prompt:
You are a customer support agent for [Brand Name].
Your goal is to resolve customer issues quickly and empathetically.

You can:
- Answer questions using the knowledge base and customer account data
- Check order status and return eligibility
- Create escalation tickets with full context when needed

You must:
- Always verify the customer's identity before accessing account data
- Escalate to a human if: the issue involves a dispute over $500, the customer asks to speak to a manager, or the query cannot be resolved within 2 exchanges
- When escalating, summarise the issue and what was already tried

Never:
- Make up policies. If unsure, search the knowledge base.
- Promise outcomes you cannot guarantee (e.g. "your refund will arrive in 2 days")

Attach all support MCP tools

Connect the MCP server with all helpdesk tools. The agent will automatically choose which tools to call based on the customer's message — checking account data, searching knowledge base, or creating tickets as needed.

Configure conversation memory

Use a consistent sessionId per customer session. SentientOne maintains full conversation history within the session — so follow-up messages like 'actually, can I just get a refund?' have full prior context.

Embed in your existing chat widget

If you already have a chat widget (Intercom, custom, etc.), route messages to SentientOne instead of or before your human queue:

// Drop into any existing chat flow
async function handleCustomerMessage(message: string, sessionId: string) {
  const res = await fetch("https://app.sentientone.ai/api/chat", {
    method: "POST",
    headers: {
      "X-Agent-Id": "YOUR_SUPPORT_AGENT_ID",
      "X-App-Key": "YOUR_APP_KEY"
    },
    body: JSON.stringify({
      message,
      sessionId  // ties full conversation history together
    })
  });
  const { reply, metadata } = await res.json();
  
  if (metadata?.escalate) {
    transferToHumanQueue(sessionId, metadata.summary);
  }
  return reply;
}

Set up human handover

When the agent creates an escalation ticket via MCP, your support team receives it in their existing helpdesk queue — already tagged with the full conversation transcript and the agent's summary of the issue.

Monitor and improve

Use SentientOne's analytics to track resolution rate, escalation rate, and average session length. As you identify common unresolved queries, update the knowledge base or add new MCP tools — no model retraining needed.