using-perplexity-for-context

Using Perplexity for Context

First-Run Setup Check

IMPORTANT: Before using perplexity tools, ensure the MCP server is built and configured:

# Check if built
ls ~/.claude/plugins/cache/brain-jar/perplexity-search/*/dist/index.js 2>/dev/null || echo "NOT_BUILT"

If NOT_BUILT, run the setup:

node ~/.claude/plugins/cache/brain-jar/perplexity-search/*/run.js &
sleep 10

Then check for API key config:

cat ~/.claude/perplexity-search/config.json 2>/dev/null || echo "NOT_CONFIGURED"

If NOT_CONFIGURED, create the config (ask user for their Perplexity API key first):

mkdir -p ~/.claude/perplexity-search
echo '{"apiKey": "USER_API_KEY_HERE", "defaultMaxResults": 5}' > ~/.claude/perplexity-search/config.json

After setup, user must restart Claude Code for MCP to register.

When to Use

Automatically invoke the perplexity_search tool in these situations:

1. Unfamiliar Territory

• Libraries, frameworks, or tools not in training data or recently updated
• New APIs, technologies, or patterns
• Example: "How do I use Bun's new test runner?"

2. Decision Points

• Choosing between architectural approaches where user preference matters
• Library selection (e.g., "Should I use Zustand or Redux?")
• Pattern choices (REST vs GraphQL, SQL vs NoSQL)

3. Learning Questions

• User asks "how does X work", "what is X", "explain Y"
• Exploratory questions about concepts or implementations
• Example: "How does React Server Components work?"

4. Preference-Dependent Choices

• Multiple valid approaches exist and user's style/preference affects the decision
• Code structure, naming conventions, testing approaches
• Example: Deciding between verbose/explicit vs concise/implicit code

5. Context Enrichment

• Answering could benefit from knowing user's background
• Technical explanations that should match user's knowledge level
• Example: Explaining advanced concepts to someone learning vs expert

How to Use

When any trigger condition is met:

1. Invoke perplexity_search tool with the query
2. Review results and citations
3. Integrate findings into response naturally
4. Include source citations in response

Do NOT announce usage unless user explicitly asks.

Subagent Pattern (Recommended)

For better token efficiency, dispatch a Haiku subagent to run the search:

Use Task tool:

• subagent_type: "general-purpose"
• model: "haiku"
• prompt: "Search for information about [topic] using the perplexity_search tool.

Return results in this format:

TL;DR: [1-2 sentence summary of the key finding]

Full Results: [Complete Perplexity response with all citations]

Keep the full response - don't over-summarize. The TL;DR is for quick scanning, but the full context is valuable for serendipitous discoveries."

Why use subagent:

• Model efficiency: Haiku handles the API call, saving Opus tokens
• TL;DR format: Quick summary at top for scanning
• Full results preserved: Serendipity matters - don't over-filter

When to skip subagent:

• Quick, simple lookups where you need immediate inline response
• When user is in the middle of a rapid back-and-forth conversation

Example output:

**TL;DR:** React Server Components render on the server and stream HTML to the client,
reducing bundle size and improving initial load time.

**Full Results:**
React Server Components (RSC) are a new paradigm for building React applications...
[full Perplexity response with citations]

Example

User: "What's the best way to handle state in React?"

[Trigger: Preference-dependent choice]
[Invoke: perplexity_search with query enriched by user profile]
[Profile context: "I prefer TypeScript, I'm learning React, I work on B2B SaaS apps"]
[Results: Personalized recommendations based on user's context]
[Response: Integrated answer with citations]

Integration with Profile

The tool automatically:

• Loads user profile from ~/.config/brain-jar/user-profile.json (shared with shared-memory plugin)
• Enriches queries with personal context
• Returns results with superior citations
• Updates profile when user mentions preferences (silent)
• Refreshes profile every 2 days from conversation history (automatic)

Note: The profile is now shared across all brain-jar plugins. Use the shared-memory plugin's learning-about-you skill for comprehensive profile management and onboarding.