Research Tool Routing

SkillAI & models

A skill for ai & models by thelobbi.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the Research Tool Routing skill

What this skill tells your AI

The instructions your AI receives, as published by thelobbi/claude in plugins/claude-code-expert/skills-old/research-routing/SKILL.md and read by ahel’s review.

Optimal routing of research tasks across Perplexity MCP, Firecrawl MCP, and Context7 MCP. Each tool has distinct strengths — use the right one for each job.

Tool Strengths at a Glance

ToolBest ForCostSpeed
Context7Library/framework docs, API referenceFreeFast (cached <5ms)
Perplexity AskKnowledge Q&A, current events, fact-checking~$0.02/queryFast
Perplexity ResearchDeep investigations, comprehensive reports~$0.10/querySlow
Firecrawl ScrapeExtract content from specific URL1 creditFast
Firecrawl MapDiscover all URLs on a site1 creditFast
Firecrawl ExtractStructured JSON from pages5-20 creditsMedium
Firecrawl AgentAutonomous multi-page research100-1500+ creditsSlow

Decision Tree

What do you need to find out?

├─ LIBRARY DOCS?
│  "How does React useEffect work?"
│  "What's the Prisma findMany API?"
│  "Next.js App Router conventions?"
│  └─→ Context7 (resolve-library-id → query-docs)
│      Free, current, version-specific, no hallucination
│
├─ GENERAL KNOWLEDGE?
│  "What is the latest in AI?"
│  "Explain event sourcing"
│  "Compare Redis vs Memcached"
│  └─→ Perplexity Ask (perplexity_ask)
│      Fast, cheap, citations included
│
├─ CURRENT EVENTS?
│  "What happened at AWS re:Invent?"
│  "Latest security vulnerability in Log4j?"
│  └─→ Perplexity Search (perplexity_search)
│      Real-time web search with ranked results
│
├─ DEEP RESEARCH?
│  "Comprehensive comparison of 5 auth solutions"
│  "Full analysis of microservice vs monolith tradeoffs"
│  └─→ Perplexity Research (perplexity_research)
│      Deep-Research model, thorough, multiple sources
│
├─ EXTRACT FROM KNOWN URL?
│  "Get the pricing from https://example.com/pricing"
│  "Extract the API reference from this page"
│  └─→ Firecrawl Scrape (firecrawl_scrape)
│      1 credit, clean markdown or JSON output
│
├─ FIND PAGES ON A SITE?
│  "What docs pages exist on docs.example.com?"
│  "Find the webhook documentation page"
│  └─→ Firecrawl Map (firecrawl_map)
│      Discover URLs, then scrape the relevant ones
│
├─ STRUCTURED DATA EXTRACTION?
│  "Get all product prices as JSON from this catalog"
│  "Extract API endpoints with parameters"
│  └─→ Firecrawl Extract (firecrawl_extract)
│      Custom JSON schema extraction
│
└─ COMPLEX MULTI-SOURCE?
   "Research X across official docs, blogs, and forums"
   └─→ Chain: Context7 → Perplexity → Firecrawl
       1. Context7: Official library docs (free)
       2. Perplexity: Broader knowledge synthesis ($0.02)
       3. Firecrawl: Extract from specific URLs found (1 credit each)

Perplexity MCP Tools

perplexity_search

  • Best for: Finding URLs, quick fact-checking, ranked results
  • Model: Sonar (fast, cheap)
  • Output: Search results with URLs, snippets, rankings

perplexity_ask

  • Best for: Direct Q&A, conversational answers with citations
  • Model: Sonar-Pro (balanced speed/quality)
  • Output: Text answer + source citations
  • Context: 128K tokens for comprehensive analysis

perplexity_research

  • Best for: Deep investigations, detailed reports
  • Model: Sonar-Deep-Research
  • Output: Comprehensive analysis with multiple sources cited
  • Cost: 3-5x more than perplexity_ask

perplexity_reason

  • Best for: Logical analysis, structured problem-solving
  • Model: Sonar-Reasoning-Pro
  • Output: Step-by-step reasoning with conclusions

Firecrawl MCP Tools

firecrawl_scrape

  • Best for: Single page content extraction
  • Cost: 1 credit (cheapest Firecrawl operation)
  • Formats: markdown, JSON, HTML, screenshot, branding
  • Tip: Use onlyMainContent: true to skip headers/footers
  • Tip: For JS-rendered pages, add waitFor: 5000

firecrawl_map

  • Best for: Discovering URLs on a site before scraping
  • Tip: Use search parameter to find specific pages
  • Workflow: Map → identify relevant URLs → Scrape each

firecrawl_search

  • Best for: Web search with optional content extraction
  • Alternative to: Perplexity search (but costs credits)
  • Tip: Prefer Perplexity for pure search; Firecrawl for search+extract

firecrawl_extract

  • Best for: Structured JSON extraction with custom schemas
  • Cost: 5-20 credits per extraction
  • Required: JSON schema defining expected output structure
  • Tip: Only use when you need structured data, not just text

firecrawl_crawl

  • Best for: Multi-page bulk extraction
  • Cost: Variable (1 credit per page crawled)
  • Warning: Can exceed token limits — set limit and maxDiscoveryDepth

firecrawl_agent

  • Cost: 100-1,500+ credits per task (UNPREDICTABLE)
  • Use: ONLY as last resort when map+scrape fails
  • Warning: Async — requires polling firecrawl_agent_status

firecrawl_browser_*

  • Best for: Interactive automation (forms, clicks, JavaScript)
  • Workflow: browser_create → browser_execute → browser_delete
  • Use: Only when scrape fails on highly interactive pages

Context7 MCP Tools

resolve-library-id

  • Purpose: Map library name to Context7 ID
  • Input: Library name (e.g., "react", "prisma", "next.js")
  • Output: Library ID (e.g., "/facebook/react")
  • Cache: 1hr (very fast on repeat calls)

query-docs

  • Purpose: Fetch documentation for a resolved library
  • Input: Library ID + specific question
  • Output: Relevant docs, API reference, examples
  • Cache: 30min for docs content
  • Tip: Be specific — "useCallback hook" not just "React"

Chaining Strategies

Strategy 1: Knowledge → Extraction

  1. Perplexity Ask: "What's the best library for X? Give me the official URL"
  2. Firecrawl Scrape: Extract clean content from the official page
  3. Benefit: Perplexity finds the best source; Firecrawl gets clean content

Strategy 2: Docs → Knowledge → Validation

  1. Context7: Get official library docs for the framework
  2. Perplexity Ask: "Common patterns for using X with Y"
  3. Context7: Verify patterns against official docs
  4. Benefit: Official docs + community knowledge, validated

Strategy 3: Map → Scrape → Synthesize

  1. Firecrawl Map: Discover all pages on a documentation site
  2. Firecrawl Scrape: Extract content from relevant pages
  3. Perplexity Ask: Synthesize findings into actionable answer
  4. Benefit: Comprehensive coverage + clean synthesis

Strategy 4: Avoid Agent Costs

  1. Perplexity Research: Get comprehensive answer with sources
  2. Only use Firecrawl Agent if structured JSON from 5+ sources is required
  3. Benefit: Cut 90% of autonomous browsing costs

Mandatory Context7 Usage

Quality audits and planners MUST use Context7.

For Quality Audits

Before auditing code:
  1. Identify libraries used in the code under review
  2. Context7: resolve-library-id for each library
  3. Context7: query-docs for current best practices
  4. Compare code against official patterns
  5. Flag deviations as audit findings with doc links

For Planning Agents

Before designing architecture:
  1. List all proposed libraries/frameworks
  2. Context7: verify API compatibility for each
  3. Context7: check for breaking changes between versions
  4. Context7: find recommended integration patterns
  5. Include doc-backed justifications in plan

For Code Review

During review:
  1. Identify library API calls in changed code
  2. Context7: verify correct API usage
  3. Context7: check for deprecated methods
  4. Flag incorrect usage with links to official docs

Cost Budget Guidelines

For a typical research task:

ApproachCostWhen to Use
Context7 onlyFreeLibrary docs, API reference
Perplexity only~$0.02Knowledge Q&A, current events
Context7 + Perplexity~$0.02Library + broader context
Perplexity + Firecrawl Scrape~$0.03Knowledge + extraction
Full chain (all 3)~$0.05Comprehensive research
Firecrawl Agent$0.08-$1.20AVOID — use chain instead

Rule: Start cheap (Context7 → Perplexity → Firecrawl Scrape). Only escalate if cheaper tools don't have the answer.

Signals

GitHub stars
21
Forks
2
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
research-routing
Source
github.com/thelobbi/claude