perplexity_research

Purpose and Scope

The perplexity_research tool provides deep, comprehensive research capabilities using the Perplexity API's sonar-deep-research model. This tool is designed for thorough analysis tasks that require detailed exploration of topics with extensive citations.

For quick questions and general queries, see perplexity_ask. For advanced logical reasoning tasks, see perplexity_reason. For direct web search without AI processing, see perplexity_search.

Tool Registration and Implementation

The perplexity_research tool is registered in the MCP server with specific configuration for deep research workflows.

Tool Registration Schema

Sources: src/server.ts298-322 src/server.ts255-271

Tool Configuration Details

The tool is registered with the following configuration:

Property	Value	Description
Tool Name	perplexity_research	Identifier used by MCP clients
Title	"Deep Research"	Human-readable name
Model	sonar-deep-research	Specialized research model
Input Type	messagesWithStripThinkingInputSchema	Messages array + optional strip_thinking
Output Type	responseOutputSchema	String response
Read-Only Hint	true	Indicates tool doesn't modify state
Open-World Hint	true	Tool accesses external information

Sources: src/server.ts298-322 src/server.ts271

Input Schema

Message Array Structure

The primary input is a messages array following the standard chat completion format:

Sources: src/server.ts255-260 src/server.ts39-56

strip_thinking Parameter

The strip_thinking parameter is optional and controls whether thinking tokens are removed from the response:

Parameter	Type	Default	Description
strip_thinking	boolean	false	If true, removes `` tags and their content from the response to save context tokens

Implementation Detail: The parameter is defined at src/server.ts262-263 and processed at src/server.ts313-316

Sources: src/server.ts262-263 src/server.ts313-316 README.md26-28

Model: sonar-deep-research

The perplexity_research tool uses the sonar-deep-research model, which is specifically optimized for comprehensive research tasks.

Model Selection Flow

Sources: src/server.ts312-321 src/server.ts62-153

Model Characteristics

The sonar-deep-research model is selected for this tool because:

1. Comprehensive Analysis: Designed for in-depth exploration of topics
2. Extended Processing: Capable of longer research workflows
3. Rich Citations: Returns detailed citation metadata
4. Thinking Tokens: May include `` tags showing reasoning process

Sources: src/server.ts316 README.md19-20

Request Processing Flow

The complete request processing pipeline for perplexity_research involves validation, API invocation, and response processing.

End-to-End Processing Pipeline