mikeng-io/agent-skills / skills/perplexity
perplexity
VerifiedOptional AI-synthesized web search via Perplexity MCP. Returns AI-grounded answers with inline citations rather than raw search results — best for synthesis-heavy questions, current consensus queries, and cross-validation of research findings. Non-blocking — falls back to other search tools if unavailable.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 6.4 KB |
SKILL.md
View on GitHub---
name: perplexity
description: Optional AI-synthesized web search via Perplexity MCP. Returns AI-grounded answers with inline citations rather than raw search results — best for synthesis-heavy questions, current consensus queries, and cross-validation of research findings. Non-blocking — falls back to other search tools if unavailable.
location: managed
context: fork
allowed-tools:
- ToolSearch
- mcp__perplexity__search
---
# Perplexity MCP — Optional AI-Synthesized Search
This skill wraps the Perplexity MCP server. Unlike raw web search tools that return a list of URLs, Perplexity returns AI-synthesized answers with inline citations — it reads the web for you and summarizes what it finds.
**Best for:** "What is the current consensus on X?", "Compare A vs B", "What changed in X since version Y?"
**Not ideal for:** Retrieving specific raw URLs, domain-specific technical documentation, anything requiring exact source control
**Setup:** Requires the Perplexity MCP server and a Perplexity API key.
```bash
# Add to MCP config (one-time setup)
claude mcp add -s user perplexity npx @perplexity-ai/mcp-server
# Then set: PERPLEXITY_API_KEY=your_key_here
# Get an API key at: https://www.perplexity.ai/settings/api
```
---
## Pre-Flight: Check Availability
```yaml
ToolSearch: "perplexity"
→ Returns: mcp__perplexity__search
→ If found: proceed to Step 1
→ If not found: return availability: "unavailable", skip — other search tools handle raw web
```
**Non-blocking:** Perplexity is a complement to web search, not a replacement. If unavailable, research continues normally with Brave Search and other tools.
---
## Step 1: Identify High-Value Query Types
Perplexity shines for these query patterns:
```yaml
high_value_use_cases:
consensus_synthesis:
description: "What does the community/industry currently think about X?"
examples:
- "What is the current consensus on Go vs Rust for systems programming?"
- "What are the most common criticisms of event sourcing in practice?"
why: Returns synthesized view across many sources, not just one opinion
comparison_analysis:
description: "Compare A vs B across multiple criteria"
examples:
- "Compare Redis vs Memcached for session storage in 2025"
- "Compare Kafka vs RabbitMQ for event streaming at scale"
why: AI synthesis better at multi-dimensional comparison than individual sources
current_state_snapshot:
description: "What is the current state of X?"
examples:
- "What is the current state of WebAssembly browser support in 2025?"
- "What authentication standards are recommended in 2025?"
why: Perplexity indexes recent content and synthesizes the current picture
cross_validation:
description: Use to validate or challenge findings from other sources
examples:
- "Are there known limitations or criticisms of [finding from web search]?"
why: Provides a second synthesis pass to surface what raw search might miss
```
**Not ideal for:**
- Retrieving specific URLs or raw source lists (use Brave Search)
- Deep technical documentation (use context7 or direct docs)
- Codebase-specific questions (use DeepWiki)
---
## Step 2: Execute Query
```
mcp__perplexity__search(query="your synthesis question here")
```
**Query construction tips:**
- Frame as a question requiring synthesis: "What is...", "How does... compare to...", "What are the tradeoffs of..."
- Include context: "in production Go services", "for startups in 2025"
- Ask for recency: "currently", "as of 2025", "latest recommendations"
- Request specific framing: "from a security perspective", "in terms of developer experience"
---
## Step 3: Process Response
Perplexity returns a synthesized answer with inline citations. Process it as:
```json
{
"source": "perplexity",
"query": "the query executed",
"answer": "Synthesized answer text with [citation] references",
"citations": [
{
"index": 1,
"url": "https://source-url",
"title": "Source title"
}
],
"credibility": "MEDIUM", // always MEDIUM — AI synthesis, not primary source
"type": "ai-synthesis",
"key_points": ["Extracted key point 1", "Extracted key point 2"]
}
```
**Credibility note:** Always tag Perplexity outputs as `credibility: MEDIUM` — the underlying sources may be HIGH, but the synthesis layer introduces potential for hallucination. Use inline citations to verify critical claims.
---
## Calling Context Integration
### When invoked by `deep-research`
Complement domain researcher queries. Run Perplexity in parallel with Brave Search for synthesis-heavy topics. Pattern:
1. Domain researcher runs Brave Search for raw source collection
2. Perplexity runs for consensus synthesis on the same topic
3. Cross-reference: does Perplexity's synthesis align with the raw sources?
4. Discrepancies become noted contradictions or gaps in research findings
### When invoked for cross-validation
After primary research is complete, run Perplexity with: "What are the main criticisms or limitations of [primary finding]?" — surfaces counter-perspectives that raw search might have missed.
### When invoked standalone
Execute 1-3 synthesis queries, return structured answer with citations. Suitable for quick orientation on an unfamiliar topic before deeper research.
---
## Output
```json
{
"skill": "perplexity",
"availability": "available | unavailable",
"queries_executed": ["list of queries"],
"results": [
{
"query": "...",
"answer": "...",
"citations": [...],
"key_points": [...]
}
],
"validation_note": "AI-synthesized answers — verify critical claims via inline citations"
}
```
If unavailable:
```json
{
"skill": "perplexity",
"availability": "unavailable",
"reason": "MCP server not configured",
"setup_hint": "claude mcp add -s user perplexity npx @perplexity-ai/mcp-server",
"alternative": "Use brave-search or web-search-prime for raw web results"
}
```
---
## Why Perplexity vs Other Search Tools?
| Tool | Returns | Best for |
|------|---------|----------|
| `brave-search` | Raw web results with URLs | Source collection, specific URL retrieval |
| `perplexity` | AI synthesis with citations | Consensus questions, comparison, current state |
| `web-search-prime` | Raw web results | General fallback search |
| `deepwiki` | Codebase wiki answers | Codebase-specific questions |
Perplexity and Brave Search are **complementary, not competing** — run both for comprehensive research coverage.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |