Recipes

Six worked examples — each one a real prompt you can paste into Claude Desktop after connecting, plus the underlying MCP tool calls so you can see what your model is actually doing.

1 — When was the last oil change on Generator 1?

construction

Pulls the most recent maintenance-task completion for an item, filtered by service type.

MCP tool call:

{
  "tool": "prism_entity_crud",
  "args": {
    "mega_tool": "entity_crud",
    "app": "spare_parts",
    "entity": "maintenance_task_completion",
    "operation": "list",
    "filters": {
      "item__name__icontains": "Generator 1",
      "task__service_type": "oil_change"
    },
    "ordering": "-completed_at",
    "page_size": 1
  }
}

Sample response shape:

{
  "success": true,
  "data": {
    "results": [{
      "id": "...",
      "completed_at": "2026-04-12T10:00:00Z",
      "performed_by": "Wilson, J.",
      "operating_hours": 1284,
      "notes": "Standard 250-hour oil change. Filter replaced."
    }],
    "pagination": { "count": 1, "page": 1 }
  }
}

Try it: In Claude Desktop, ask: "When was the last oil change on Generator 1?"

2 — Create a work order for the fire pump

construction

Creates a new work order for a maintenance item. Demonstrates a write operation with required fields.

MCP tool call:

{
  "tool": "prism_entity_crud",
  "args": {
    "mega_tool": "entity_crud",
    "app": "spare_parts",
    "entity": "work_order",
    "operation": "create",
    "data": {
      "title": "Fire pump inspection",
      "item_id": "<fire-pump-uuid>",
      "priority": "medium",
      "description": "Quarterly inspection per NFPA 25.",
      "due_date": "2026-05-19"
    }
  }
}

Try it: "Open a work order to inspect the fire pump next week."

3 — Show overdue invoices and email a summary

finance

Demonstrates tool chaining: query overdue invoices, then dispatch a templated email.

Step 1 — Query overdue invoices:

{
  "tool": "prism_entity_crud",
  "args": {
    "mega_tool": "entity_crud",
    "app": "finance",
    "entity": "invoice",
    "operation": "list",
    "filters": {
      "status": "open",
      "due_date__lt": "today"
    },
    "ordering": "due_date",
    "page_size": 50
  }
}

Step 2 — Email a summary to the finance inbox:

{
  "tool": "prism_notification_send",
  "args": {
    "mega_tool": "notification_send",
    "app": "core",
    "entity": "artifact_email",
    "operation": "send",
    "to": ["[email protected]"],
    "subject": "Overdue invoices — week of {{week_of}}",
    "body": "Top 10 overdue:\n\n{{rendered_table}}\n\nFull list attached.",
    "attachments": [{"name": "overdue.csv", "content_b64": "..."}]
  }
}

Try it: "Show me all overdue invoices, then email finance a summary."

4 — Search documents for "GIE turbocharger" and read the top result

documents

Combines semantic-search with a follow-up read for full-text retrieval.

Step 1 — Search:

{
  "tool": "prism_document_op",
  "args": {
    "mega_tool": "document_op",
    "app": "documents",
    "entity": "document",
    "operation": "search",
    "query": "GIE turbocharger",
    "limit": 5
  }
}

Step 2 — Read the top hit:

{
  "tool": "prism_entity_crud",
  "args": {
    "mega_tool": "entity_crud",
    "app": "documents",
    "entity": "document",
    "operation": "read",
    "id": "<top-result-uuid>"
  }
}

Try it: "Find documents about GIE turbocharger and summarize the top result."

5 — Spawn a long-running agent to audit spare-parts inventory

agent

Most MCP servers expose CRUD. Cognethics also exposes agentic primitives — spawn a long-running agent with its own budget, persona, and authority level. The agent runs on the platform, delegates to sub-agents as needed, and reports back asynchronously when the mission completes. The calling model gets back a mission ID it can poll, not a blocking response.

Single spawn call (the tease):

{
  "tool": "pj_v3_spawn_agent",
  "args": {
    "mission_intent": "Audit spare-parts inventory: identify items with no service history in 12+ months and items with stock below reorder threshold.",
    "persona": "DocIntel",
    "authority_level": "delegate",
    "budget_usd": 5.0
  }
}

Full walkthrough: see /recipes/spawn-agent/ for the end-to-end agent-OS round-trip — mission → spawn → budget → wait → fold → audit, with the full HTTP shape for each call.

Want more? Browse the full handler catalog — every handler can be combined into custom recipes.