Deep Research Engine

The BPAI MCP Server includes a deep research engine that performs real web searches before content generation. This produces articles grounded in current facts, statistics, and verified sources.

Auto-Research (Built Into Generation)

You don't need to call research_topic manually. Research runs automatically inside generate_article and generate_batch by default.

When you run:

Generate an article about "best CRM software for small businesses"

This is what actually happens:

1. 🔬 Research Phase (Perplexity Sonar Pro)
   ├── Live web search for "best CRM software for small businesses"
   ├── Collects 5-15 source URLs with statistics
   └── Returns structured findings

2. 📝 Generation Phase
   ├── Research data injected as context
   ├── Knowledge base injected
   ├── Internal links injected
   └── AI writes using real-world data + citations

Controlling Research

Want Set
Research + Generate (default) Just call generate_article normally
Skip research auto_research: false
Use your own research Provide research_context string
Research + Generate + Humanize humanize: true

Research Provider Fallback

Auto-research tries providers in this order until one works:

  1. Perplexity (preferred, best citations)
  2. Gemini (Google Search grounding)
  3. OpenAI (web search preview)
  4. Grok (live web + X/Twitter)
  5. DeepSeek (URL extraction from output)

It uses whichever provider has an API key configured. If research fails for any reason, the article still generates (just without research data).

research_topic (Standalone)

You can still call research_topic directly if you want to research a topic without generating an article. Useful for gathering intel, comparing data, or feeding research into a custom workflow.

Parameters

Parameter Type Required Default Description
topic string ✅ — Topic or keyword to research
provider string ❌ perplexity perplexity, gemini, deepseek, grok, or openai
model string ❌ Provider default Model override (e.g., sonar-deep-research)
focus string ❌ — Focus area (e.g., "pricing", "competitors")

Example

Research "AI chip market trends 2026" with focus on 
"pricing and supply chain for NVIDIA Blackwell vs AMD MI400"

Response

{
  "status": "success",
  "topic": "AI chip market trends 2026",
  "provider": "perplexity",
  "model": "sonar-pro",
  "research": "The AI chip market is projected to reach $128 billion by 2026...",
  "citations": [
    "https://www.reuters.com/technology/ai-chip-market-2026",
    "https://www.tomshardware.com/nvidia-blackwell-update",
    "https://semianalysis.com/amd-mi400-deep-dive"
  ],
  "citation_count": 3
}

Research Providers

Provider Default Model How It Searches Best For
Perplexity sonar-pro Native web search with verified citations Statistics, product comparisons, current events
Gemini gemini-3-flash-preview Google Search grounding tool Google-indexed content, academic sources
OpenAI gpt-4o-search-preview web_search_options with URL annotations Broad web coverage
Grok grok-3 web_search tool with return_citations Trending topics, X/Twitter data, breaking news
DeepSeek deepseek-chat No native search; extracts URLs from output Technical topics, multi-step reasoning

Gemini Search Details

Gemini uses Google's google_search grounding tool. It has automatic retry with exponential backoff on transient 503 errors, and falls back through models: gemini-3-flash-preview → gemini-2.5-flash → gemini-2.5-pro.

Research Cost Estimation

Provider Cost per Query 50 Articles 500 Articles
Perplexity (Sonar Pro) ~$0.05 ~$2.50 ~$25
Gemini ~$0.02 ~$1.00 ~$10
DeepSeek ~$0.01 ~$0.50 ~$5
Grok ~$0.04 ~$2.00 ~$20
OpenAI ~$0.03 ~$1.50 ~$15

Costs are approximate and depend on query complexity and response length.

Next Steps