Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/perplexityai/modelcontextprotocol/llms.txt

Use this file to discover all available pages before exploring further.

Overview

The perplexity_research tool performs comprehensive, in-depth research using the Sonar Deep Research model. It analyzes multiple sources to provide thorough, well-researched answers with extensive citations.
This tool is significantly slower than other tools (typically 30+ seconds). Use it only when you need comprehensive, multi-source analysis. For quick questions, use perplexity_ask instead.
This tool uses the sonar-deep-research model with streaming enabled for real-time progress updates during long research operations.

Use Cases

  • Literature reviews and academic research
  • Comprehensive topic overviews
  • Investigative journalism queries
  • Multi-perspective analysis
  • In-depth market research
  • Complex technical documentation

Parameters

messages
array
required
Array of conversation messages in OpenAI format.Each message must have:
  • role: One of "system", "user", or "assistant"
  • content: The message text (string)
Example:
[
  { "role": "user", "content": "Comprehensive analysis of renewable energy trends" }
]
strip_thinking
boolean
default:false
If true, removes <think>...</think> tags and their content from the response.The model may include internal reasoning in <think> tags. Stripping these saves context tokens in long conversations.Default: false (shows thinking process)Example: true
reasoning_effort
enum
Controls the depth of deep research reasoning.Options: "minimal", "low", "medium", "high"
  • "minimal": Fastest, less thorough analysis
  • "low": Balanced speed and depth
  • "medium": Comprehensive analysis
  • "high": Most thorough, slowest
Higher values produce more detailed analysis but take longer.Example: "high"

Response Format

The tool returns a detailed, AI-generated research report with extensive citations:
response
string
Comprehensive research response with:
  • In-depth analysis from multiple sources
  • Numbered citation references [1], [2], etc.
  • Optional <think>...</think> tags (if strip_thinking: false)
  • Citations list appended at the end
Response Structure:
[Optional: <think>internal reasoning</think>]

Detailed research content with inline citations[1][2]...

Citations:
[1] https://source1.com
[2] https://source2.com
...

Example Response

<think>
This query requires analyzing renewable energy trends across multiple sectors:
solar, wind, hydro, and emerging technologies. I'll need to examine recent
data, policy changes, and technological advancements.
</think>

Renewable Energy Trends: Comprehensive Analysis

The renewable energy sector has experienced unprecedented growth in 2025-2026,
with global capacity additions reaching 400 GW annually[1]. This represents a
45% increase from the previous year[2].

Solar Power Developments:
Solar photovoltaic installations dominated growth, accounting for 60% of new
renewable capacity[3]. Perovskite-silicon tandem cells achieved commercial
viability with 33% efficiency rates[4], marking a significant breakthrough
from traditional silicon cells' 22% efficiency[5].

Wind Energy Advancements:
Offshore wind capacity expanded by 35 GW globally[6], with floating turbines
enabling deployment in deeper waters[7]. The average turbine size increased
to 15 MW, up from 8 MW in 2020[8].

[... extensive research continues ...]

Citations:
[1] https://iea.org/reports/renewable-energy-2026
[2] https://irena.org/publications/2026/capacity-statistics
[3] https://solar.org/growth-report-2026
[4] https://nature.com/articles/perovskite-tandem
[5] https://nrel.gov/pv/efficiency-records
[6] https://gwec.net/offshore-wind-report
[7] https://energy.gov/floating-turbines
[8] https://windeurope.org/turbine-statistics

Usage Examples

const result = await use_mcp_tool(
  "perplexity",
  "perplexity_research",
  {
    messages: [
      {
        role: "user",
        content: "Comprehensive overview of CRISPR gene editing applications"
      }
    ]
  }
);

Common Patterns

Academic Literature Review

Produce comprehensive research summaries:
const review = await use_mcp_tool(
  "perplexity",
  "perplexity_research",
  {
    messages: [
      {
        role: "system",
        content: "Structure the response as an academic literature review with: Introduction, Methodology, Key Findings, Discussion, and Conclusion."
      },
      {
        role: "user",
        content: "Recent advances in neural network pruning techniques"
      }
    ],
    reasoning_effort: "high"
  }
);

Market Research

Analyze industry trends and competitive landscapes:
const market = await use_mcp_tool(
  "perplexity",
  "perplexity_research",
  {
    messages: [
      {
        role: "user",
        content: "Global electric vehicle market analysis: trends, key players, and future outlook"
      }
    ],
    reasoning_effort: "medium"
  }
);

Technical Deep Dive

Investigate complex technical topics:
const technical = await use_mcp_tool(
  "perplexity",
  "perplexity_research",
  {
    messages: [
      {
        role: "user",
        content: "How do transformer models handle long-context sequences? Cover architectural innovations, attention mechanisms, and computational efficiency."
      }
    ],
    reasoning_effort: "high",
    strip_thinking: false  // Keep reasoning visible
  }
);

Advanced Usage

Multi-turn Research Conversation

// Initial broad research
const initial = await use_mcp_tool(
  "perplexity",
  "perplexity_research",
  {
    messages: [
      { role: "user", content: "Overview of blockchain consensus mechanisms" }
    ]
  }
);

// Follow-up deep dive
const followup = await use_mcp_tool(
  "perplexity",
  "perplexity_research",
  {
    messages: [
      { role: "user", content: "Overview of blockchain consensus mechanisms" },
      { role: "assistant", content: initial.response },
      { role: "user", content: "Focus on proof-of-stake vs proof-of-work energy consumption" }
    ],
    reasoning_effort: "high"
  }
);

Token Optimization

For long conversations, strip thinking to save tokens:
const optimized = await use_mcp_tool(
  "perplexity",
  "perplexity_research",
  {
    messages: [
      { role: "user", content: "Comprehensive analysis of protein folding" }
    ],
    strip_thinking: true,  // Save context tokens
    reasoning_effort: "medium"
  }
);

Tips and Best Practices

Use for complex topics only: This tool is designed for queries that require synthesizing information from many sources. For simple questions, use perplexity_ask.
Be patient: Research queries typically take 30-60 seconds. Increase PERPLEXITY_TIMEOUT_MS if needed (default: 300000ms = 5 minutes).
Adjust reasoning effort: Use "high" for critical research, "medium" for balanced speed/depth, and "low" or "minimal" when time is limited.
Strip thinking in production: Enable strip_thinking: true for end-user applications to provide cleaner responses and save tokens.
Provide clear context: The more specific your query, the better the research quality. Include specific aspects you want covered.
Timeout considerations: Very complex research with reasoning_effort: "high" may exceed the default 5-minute timeout. Set PERPLEXITY_TIMEOUT_MS=600000 (10 minutes) for such queries.
The tool uses streaming to provide real-time progress updates. You’ll receive partial results as the research progresses, rather than waiting for the entire response.

Understanding Thinking Tags

When strip_thinking: false (default), responses may include <think>...</think> tags:
<think>
This query requires examining:
1. Historical context of the technology
2. Current state-of-the-art implementations
3. Future projections and challenges
I'll structure this as a chronological analysis.
</think>

[Main response content...]
Thinking tags show:
  • Research strategy and planning
  • Source evaluation reasoning
  • Analytical approach
  • Information synthesis logic
Set strip_thinking: true to remove these tags and save ~10-20% of response tokens.

Comparison with Other Tools

Featureperplexity_researchperplexity_askperplexity_reason
Modelsonar-deep-researchsonar-prosonar-reasoning-pro
SpeedVery slow (30s+)Fast (< 5s)Medium (5-15s)
SourcesMany (10+)Few (3-5)Moderate (5-8)
DepthComprehensiveStandardAnalytical
Best ForLiterature reviewsQuick Q&ALogic problems
Thinking TagsYes (optional)NoYes (optional)
Reasoning EffortAdjustableN/AN/A
FiltersNoYes (recency, domain, context)Yes (recency, domain, context)
perplexity_research is the only tool that supports reasoning_effort parameter and uses streaming by default.

Error Handling

Common errors and solutions:
  • Timeout errors: Increase PERPLEXITY_TIMEOUT_MS environment variable
  • Invalid messages array: Ensure each message has role and content fields
  • Invalid reasoning effort: Must be minimal, low, medium, or high
  • Network errors: Check API key and connectivity
  • Stream parsing errors: Usually transient, retry the request

Performance Optimization

Reduce Timeout

Use reasoning_effort: "low" or "minimal" for faster results

Save Tokens

Enable strip_thinking: true to reduce response size by 10-20%

Cache Results

Store research results locally to avoid re-running expensive queries

Parallel Research

Run multiple focused queries in parallel instead of one broad query

Ask Tool

Use for quick questions that don’t need deep research

Reason Tool

Use for logical analysis and step-by-step reasoning

Search Tool

Use to find sources before conducting research

Tools Overview

Compare all available tools and their capabilities

Build docs developers (and LLMs) love