Use Cases

A company has an MCP server that exposes order management tools. They create an agent in the SentientOne platform that instructs the LLM to use those tools to fetch order details, delivery status, and product information.

You are an Order Management Assistant connected to the
company MCP server. You have access to:
  get_order(order_id, user_id)
  get_delivery_status(order_id)
  get_order_products(order_id)
When asked about an order, extract the OrderId and UserId,
call the appropriate tools, and return a JSON summary with
status, items, prices, delivery estimate, and tracking number.

1. 1

   Create the agent in the platform UI

   Pick Custom (or an Order template if available), paste the system prompt above, set output type to JSON, attach your MCP server, and save.
2. 2

   Call the agent from your application

   bash

   curl -N -X POST https://api.sentientone.ai/v1/chat/stream \
     -H "Content-Type: application/json" \
     -H "X-Api-Key: sk-so-your_api_key_here" \
     -H "X-Agent-Id: ORDER_AGENT_UUID" \
     -d '{
       "message": "Get full order details including delivery and products for OrderId: ORD-78923 and UserId: USR-4412"
     }'

3. 3

   Receive the streamed response (SSE)

   text

   data: {"type":"meta","conversation_id":"conv-uuid-...","trace_id":"trace-uuid-..."}
   
   data: {"type":"sources","sources":[{"index":0,"id":"src-...","title":"Order ORD-78923","url":null,"source_type":"mcp","score":0.91,"snippet":"status: shipped · carrier: FedEx"}]}
   
   data: {"type":"delta","content":"{\"order_id\":\"ORD-78923\","}
   data: {"type":"delta","content":"\"status\":\"shipped\",\"total\":115.97,"}
   data: {"type":"delta","content":"\"delivery\":{\"tracking\":\"FX-998877\"}}"}
   
   data: {"type":"done","conversation_id":"conv-uuid-...","trace_id":"trace-uuid-..."}

   The full answer is the concatenation of every delta event's content in order. For a JSON output agent, concatenate all deltas first, then JSON.parse() the assembled string.

A separate agent focused purely on product catalog queries — using Anthropic's Claude with different MCP tools and a different system prompt. Same caller, different X-Agent-Id.

You are a Product Catalog Assistant. You have access to:
  search_products(query, category?, limit?)
  get_product(product_id)
  get_recommendations(product_id, limit?)
Return comprehensive details including name, price,
availability, key specs, and related products.

curl -N -X POST https://api.sentientone.ai/v1/chat/stream \
  -H "Content-Type: application/json" \
  -H "X-Api-Key: sk-so-your_api_key_here" \
  -H "X-Agent-Id: PRODUCT_AGENT_UUID" \
  -d '{ "message": "Show me the details for product PRD-2210 and suggest similar items" }'

The response streams back as Server-Sent Events — a meta event, an optional sources event, a series of delta events whose content you concatenate to form the full answer, and a final done event.

The recommended layout for companies integrating SentientOne into their stack:

• 1. Create agentsIn the SentientOne platform, configure agents with system prompts, models, and LLM keys for each use case.
• 2. Copy your API keyFrom Settings → API Key (sk-so-…).
• 3. Copy your Agent IDFrom Agents → Agent ID chip.
• 4. Call from your appPOST https://api.sentientone.ai/v1/chat/stream with X-Api-Key and X-Agent-Id headers, body { "message": "…" }.
• 5. Read the SSE streamConcatenate every delta event's content for the full AI response; the sources event lists any retrieved or MCP-tool context, and done carries the final conversation_id.