Research Pipeline

Build an automated research workflow that fetches information, summarizes findings, and extracts key insights.

Difficulty: Intermediate Time: 20 minutes

What You'll Build

A flow that:

• Takes a research topic as input
• Searches for relevant information
• Summarizes findings
• Extracts actionable insights
• Outputs a structured report

TBD: Replace with screenshot of the research pipeline in the flow editor

Prerequisites

• Sciorex installed
• Understanding of Flows
• (Optional) Web search MCP server configured
• (Optional) Research MCP server enabled for academic paper search

Overview

┌─────────┐    ┌──────────┐    ┌───────────┐    ┌────────────┐
│  Input   │───▶│  Search  │───▶│ Summarize │───▶│  Extract   │
│  Topic   │    │  Agent   │    │   Agent   │    │  Insights  │
└─────────┘    └──────────┘    └───────────┘    └────────────┘
                                                       │
                                                       ▼
                                               ┌────────────┐
                                               │    End     │
                                               └────────────┘

Step 1: Create the Agents

Search Agent

name: Research Searcher
description: Searches for information on a topic

systemPrompt: |
  You are a research assistant. Given a topic:

  1. Use web search to find relevant, authoritative sources
  2. Focus on recent information (last 2 years preferred)
  3. Gather at least 5 different sources
  4. For each source, note:
     - Title and URL
     - Key points
     - Publication date
     - Credibility assessment

  When the sciorex-research MCP server is available, also use
  sciorex_search_papers to search academic databases for peer-reviewed
  sources. This searches across Semantic Scholar, OpenAlex, CrossRef,
  arXiv, PubMed, and DBLP simultaneously.

  Return structured data that can be processed further.

model: claude-sonnet-5-0
mcpServers:
  - sciorex-research
allowedTools:
  - tool: WebSearch
    allowed: true
  - tool: WebFetch
    allowed: true

Summarizer Agent

name: Research Summarizer
description: Summarizes research findings

systemPrompt: |
  You are an expert at synthesizing research. Given multiple sources:

  1. Identify common themes and consensus views
  2. Note any contradictions or debates
  3. Highlight the most important findings
  4. Organize by relevance, not source

  Write a clear, comprehensive summary that someone unfamiliar
  with the topic could understand.

model: claude-sonnet-5-0
thinkingLevel: think

Insight Extractor Agent

name: Insight Extractor
description: Extracts actionable insights from research

systemPrompt: |
  You are an analyst who extracts actionable insights. Given research:

  1. Identify key takeaways
  2. Note practical applications
  3. Highlight risks or concerns
  4. Suggest next steps or areas for deeper research

  Format as a structured report with clear sections.

model: claude-sonnet-5-0
thinkingLevel: think-hard

outputSchema:
  type: object
  properties:
    keyTakeaways:
      type: array
      items:
        type: string
    applications:
      type: array
      items:
        type: string
    risks:
      type: array
      items:
        type: string
    nextSteps:
      type: array
      items:
        type: string

Step 2: Build the Flow

1. Go to Flows → + New Flow
2. Name it "Research Pipeline"

Flow Configuration

{
  "name": "Research Pipeline",
  "description": "Automated research workflow",
  "nodes": [
    {
      "id": "trigger",
      "type": "trigger",
      "label": "Research Input",
      "config": {},
      "position": { "x": 0, "y": 0 },
      "triggerType": "manual",
      "triggerConfig": {
        "inputSchema": {
          "type": "object",
          "properties": {
            "topic": { "type": "string", "description": "Research topic to investigate" },
            "depth": { "type": "string", "description": "Research depth: quick, standard, deep" }
          },
          "required": ["topic"]
        }
      }
    },
    {
      "id": "searcher",
      "type": "agent",
      "label": "Search",
      "config": {},
      "position": { "x": 250, "y": 0 },
      "agentId": "research-searcher",
      "inputMapping": {
        "topic": "trigger.input.topic",
        "depth": "trigger.input.depth"
      },
      "outputMapping": {}
    },
    {
      "id": "summarizer",
      "type": "agent",
      "label": "Summarize",
      "config": {},
      "position": { "x": 500, "y": 0 },
      "agentId": "research-summarizer",
      "inputMapping": {
        "topic": "trigger.input.topic",
        "sources": "nodes.searcher.output"
      },
      "outputMapping": {}
    },
    {
      "id": "extractor",
      "type": "agent",
      "label": "Extract Insights",
      "config": {},
      "position": { "x": 750, "y": 0 },
      "agentId": "insight-extractor",
      "inputMapping": {
        "summary": "nodes.summarizer.output",
        "sources": "nodes.searcher.output"
      },
      "outputMapping": {}
    },
    {
      "id": "end",
      "type": "end",
      "label": "Complete",
      "config": {},
      "position": { "x": 1000, "y": 0 },
      "endType": "success"
    }
  ],
  "edges": [
    { "id": "e1", "source": "trigger", "target": "searcher" },
    { "id": "e2", "source": "searcher", "target": "summarizer" },
    { "id": "e3", "source": "summarizer", "target": "extractor" },
    { "id": "e4", "source": "extractor", "target": "end" }
  ],
  "variables": {},
  "maxIterations": 5
}

Step 3: Run the Pipeline

1. Click Run in the flow editor
2. Enter your research topic:

   {
     "topic": "Current state of AI code assistants in 2025",
     "depth": "standard"
   }

3. Watch the execution progress through each stage

Example Output

{
  "keyTakeaways": [
    "AI code assistants have reached mainstream adoption with 70%+ developer usage",
    "Code quality improvements of 15-30% reported across studies",
    "Security concerns remain around training data and generated code"
  ],
  "applications": [
    "Automated code review and bug detection",
    "Documentation generation",
    "Test case creation",
    "Legacy code modernization"
  ],
  "risks": [
    "Over-reliance leading to skill atrophy",
    "Potential for introducing subtle bugs",
    "License and copyright considerations"
  ],
  "nextSteps": [
    "Evaluate specific tools for your tech stack",
    "Establish code review processes for AI-generated code",
    "Monitor emerging research on AI code quality"
  ]
}

Using Research MCP Tools

This pipeline becomes more powerful when agents have access to the Research MCP server. Enable sciorex-research in the agent's MCP servers to give it access to:

• sciorex_search_papers — Search across 6 academic databases (Semantic Scholar, OpenAlex, CrossRef, arXiv, PubMed, DBLP)
• sciorex_add_reference — Save papers to the reference library for future citation
• sciorex_compile_latex — Generate formatted summaries as LaTeX documents
• sciorex_annotate_pdf — Highlight key findings in downloaded PDFs

This lets the agent go beyond web search to use structured academic APIs with citation-aware results. When the Search Agent finds relevant papers via sciorex_search_papers, it can automatically save them to your reference library with sciorex_add_reference, building a curated bibliography as part of the pipeline.

Variations

Add Parallel Sources

Modify the flow to search multiple source types in parallel:

                  ┌── Academic Search ──┐
Input ── Parallel ┼── News Search ──────┼── Merge ── Summarize ── End
                  └── Technical Blogs ──┘

Use a Parallel node with branches pointing to three search agent nodes, then a Merge node with mergeStrategy: "all" and sourceNodes listing all three.

Add Fact Checking

Insert a verification step:

Summarize → Fact Checker → Extract Insights → End

Export to Ticket

Add a Ticket Action node at the end:

action: "create"
actionParams:
  title: "Research: trigger.input.topic"
  description: "nodes.extractor.output"

Use link_current_flow_run to link the execution to the ticket for full traceability.

Tips

• Start narrow: Begin with specific topics before broad ones
• Check sources: Review the searcher's sources for quality
• Iterate: Run multiple times with refined prompts
• Save outputs: Link to tickets for future reference
• Enable Research MCP: For academic topics, enable the sciorex-research MCP server to access structured paper databases alongside general web search
• Use debug mode: Step through nodes to verify data quality at each stage

Related

• Flow Editor Overview
• Agent Outputs
• MCP Servers
• Research MCP Tools