Perplexity Search Strategy Guide

This skill guides optimal use of the four Perplexity MCP tools. It does NOT provide search capability itself — the perplexity MCP server (package: @perplexity-ai/mcp-server) provides the actual tools.

MCP server name: perplexity → tools appear as mcp__perplexity__perplexity_search, etc.

Tool Selection

You have four Perplexity tools. Pick the right one:

Tool	Use When	Speed	Cost
perplexity_search	Finding URLs, checking news, discovering sources	Fast	Lowest
perplexity_ask	Quick factual questions, summaries, explanations	Fast	Low
perplexity_reason	Complex analysis, comparisons, step-by-step logic	Medium	Medium
perplexity_research	Deep literature reviews, comprehensive overviews	Slow (30s+)	Highest

Decision Flow

Need specific URLs or source discovery?
  → perplexity_search

Quick factual answer or summary?
  → perplexity_ask

Needs step-by-step reasoning or comparison?
  → perplexity_reason

Needs deep multi-source synthesis (and time isn't critical)?
  → perplexity_research

Default to perplexity_ask — it handles 80% of queries well. Only escalate when you genuinely need reasoning or deep research.

Tool Parameters

perplexity_search

Returns ranked results (title, URL, snippet, date) — no AI synthesis.

• query (required): Search query string
• max_results: 1-20 (default: 10)
• max_tokens_per_page: 256-2048 (default: 1024) — tokens to extract per page
• country: ISO country code for regional results

perplexity_ask

AI-generated answer with citations. Fastest AI tool.

• messages (required): Conversation messages array
• search_recency_filter: hour | day | week | month | year
• search_domain_filter: Restrict to specific domains, prefix - to exclude
• search_context_size: low (default, fastest) | medium | high

perplexity_reason

Step-by-step reasoning with web grounding.

• messages (required): Conversation messages array
• search_recency_filter: Same as ask
• search_domain_filter: Same as ask
• search_context_size: Same as ask
• strip_thinking: Remove <think> tags from response (saves context tokens)

perplexity_research

Deep multi-source research. 30+ seconds per query.

• messages (required): Conversation messages array
• reasoning_effort: minimal | low | medium | high
• strip_thinking: Remove <think> tags from response

Query Crafting

Structure Every Query

Good queries have four components:

1. Topic: Main subject
2. Scope: Specific aspect
3. Context: Time frame, domain, constraints
4. Output: What kind of answer you need

Bad: "Tell me about cancer treatment" Good: "What are the latest clinical trial results for CAR-T cell therapy in treating B-cell lymphoma published in 2024-2025?"

Use Recency Filters

The search_recency_filter parameter is powerful — use it:

• Breaking news → hour
• Today's developments → day
• This week's updates → week
• Recent research → month or year

Use Domain Filters

Restrict to authoritative sources with search_domain_filter:

• Scientific: ["arxiv.org", "pubmed.ncbi.nlm.nih.gov", "nature.com", "science.org"]
• Technical: ["github.com", "stackoverflow.com", "docs.python.org"]
• News: ["reuters.com", "apnews.com"]
• Exclude noise: ["-reddit.com", "-quora.com"]

Use search_context_size Strategically

• low (default): Fast, good enough for most queries
• medium: When initial results seem thin
• high: For comprehensive topics where you need broad coverage

Domain-Specific Patterns

Scientific Literature

{
  "messages": [
    {
      "role": "user",
      "content": "What does recent research say about the role of gut microbiome in Parkinson's disease? Focus on peer-reviewed studies with specific bacterial species identified."
    }
  ],
  "search_recency_filter": "year",
  "search_domain_filter": ["pubmed.ncbi.nlm.nih.gov", "nature.com", "cell.com", "science.org"],
  "search_context_size": "high"
}

Use precise terminology: "randomized controlled trial" not "study", "meta-analysis" not "review", specific gene/protein names not general descriptions.

Technical Documentation

{
  "messages": [
    {
      "role": "user",
      "content": "How to implement real-time data streaming from Kafka to PostgreSQL using Python with backpressure handling and exactly-once semantics?"
    }
  ],
  "search_domain_filter": ["docs.confluent.io", "github.com", "stackoverflow.com"]
}

Include framework versions, specific APIs, and implementation constraints.

Current Events / News

{
  "messages": [
    {
      "role": "user",
      "content": "What was announced regarding AI regulation at the EU parliament this week?"
    }
  ],
  "search_recency_filter": "week",
  "search_context_size": "medium"
}

Comparative Analysis

Use perplexity_reason for comparisons — it shows its thinking:

{
  "messages": [
    {
      "role": "user",
      "content": "Compare PyTorch versus TensorFlow for implementing transformer models in terms of ease of use, performance, and ecosystem support. Include benchmarks from recent studies."
    }
  ],
  "search_context_size": "high",
  "strip_thinking": true
}

Deep Research

Use perplexity_research only when you need comprehensive multi-source synthesis:

{
  "messages": [
    {
      "role": "user",
      "content": "Conduct a comprehensive analysis of competing CAR-T cell therapy approaches including mechanism differences, clinical outcomes, cost-effectiveness, and manufacturing challenges."
    }
  ],
  "reasoning_effort": "high"
}

Common Mistakes

Mistake	Fix
Using perplexity_research for simple facts	Use perplexity_ask — 10x faster, much cheaper
Vague queries ("tell me about AI")	Be specific: topic + scope + context + output
Not using recency filters	Add search_recency_filter when freshness matters
Not using domain filters	Restrict to authoritative sources for quality
Asking multiple unrelated questions in one query	One focused question per tool call
Using search_context_size: high by default	Start with low, escalate only if results are thin
Forgetting strip_thinking on reason/research	Set true to save context window tokens

Iterative Search Workflow

For comprehensive research, chain tools:

1. Discover (perplexity_search): Find key sources and URLs
2. Summarize (perplexity_ask): Get overview from those sources
3. Analyze (perplexity_reason): Compare approaches or reason through findings
4. Deep dive (perplexity_research): Only if steps 1-3 leave gaps

Don't jump to perplexity_research first — build understanding incrementally.

Reference

See references/search_strategies.md for detailed query templates by domain (biomedical, computational, clinical, chemistry).