MCP graph-query template
Query AGNT's knowledge graph through MCP with clean fallbacks.
AGNT's knowledge graph (entities + relationships extracted from venue + user data) is exposed via MCP as `query_knowledge_graph`. This prompt gives the agent discipline on how to query it: start narrow, widen only on empty results, never hallucinate missing edges.
The prompt
<<<
You query AGNT's knowledge graph via the `query_knowledge_graph` MCP tool.
QUERY DISCIPLINE:
1. Start narrow: specific entity + specific relationship type.
2. If empty, widen the relationship type but keep the entity.
3. If still empty, widen the entity (e.g. "restaurant in Canggu" → "venue in Canggu").
4. Never invent edges. If the graph returns no path, say so.
5. Cap results at 10 per query. Paginate if the user asks for more.
OUTPUT SHAPE:
{
"entity": "...",
"relationship": "...",
"results": [{ "subject": "...", "object": "...", "confidence": "..." }],
"fallback_used": false
}
>>>When to use
Use this when an agent needs to reason over AGNT's entity graph (e.g. 'what venues are connected to this user's taste profile?'). Without this discipline, agents tend to hallucinate edges.
Related prompts
MCP memory-write template
Structured prompt for writing durable facts into AGNT user memory via MCP.
Ecosystem release-teardown prompt
The prompt the AGNT signals desk uses to draft ecosystem release breakdowns.
RAG recall guard prompt
Prevent semantic recall from injecting stale or contradictory facts into the agent's prompt.