pinecone-assistant-paid-plan

Pinecone Assistant MCP - Paid Plan Skill

assistant_chat is a sub-agent delegation mechanism. Pinecone internally retrieves context from the knowledge base, feeds it to the configured LLM, and returns a synthesized, citation-backed answer. Claude receives only the compact result (~500–2000 tokens) rather than raw document chunks.

evaluate_answer is a quality gate. Given a question, a generated answer, and a ground truth answer, it scores correctness (precision), completeness (recall), and alignment (harmonic mean) — and returns per-fact entailment reasoning.

Primary tools: assistant_chat, evaluate_answer Primary prompt: delegated_research

Available Tools

Tool Selection Decision Tree

Choose based on the task:

User wants to find information
├── Need synthesized answer + citations (let Pinecone do the work)
│   └── assistant_chat (context_options={"top_k": 5, "snippet_size": 2048})
│       └── delegated_research prompt for guided workflow
│
├── Need to validate answer quality against known ground truth
│   └── evaluate_answer(question, answer, ground_truth_answer)
│       ├── Scores: correctness (precision), completeness (recall), alignment (F1)
│       └── Use after assistant_chat in agentic/evaluation workflows
│
├── Claude's context window under pressure
│   └── assistant_chat (delegation preserves context window)
│       └── Chain independent calls without history
│
├── Need raw chunks to synthesize yourself / control the output
│   ├── Single topic → assistant_context (top_k=3-5)
│   └── Multi-angle → assistant_strategic_multi_search_context
│
└── Need AI to reason across many search patterns
    └── assistant_strategic_multi_search_chat