anombyte93/deep-research-mcp

Deep Research MCP Server

Multi-agent research system with configurable depth for Claude Desktop

Deep Research MCP is an exportable, standalone MCP server that implements parallel multi-agent research with configurable depth modes. Inspired by the /deep-research slash command pattern, it addresses key industry gaps identified in 2024-2025 AI tool research.

Key Features

• Multi-Agent Parallel Research: 2-10 agents simultaneously (not sequential)
• Configurable Depth Modes: Quick (5min, ~$0.30) | Standard (15min, ~$1.00) | Deep (45min, ~$3.00)
• Provider Flexibility: Works with Anthropic Claude, OpenAI GPT, or Google Gemini
• Evidence-Based: Confidence scores + citations for all findings
• Industry Gap-Filling: Addresses context rot, verification overhead, ROI measurement
• Export-Ready: Standalone server, anyone can install with their API key

Quick Start (5 Minutes)

See for detailed setup instructions.

# 1. Clone and install
git clone https://github.com/your-org/deep-research-mcp.git
cd deep-research-mcp
npm install

# 2. Configure API key
cp .env.example .env
# Edit .env and add your ANTHROPIC_API_KEY

# 3. Build
npm run build

# 4. Test
npm run dev

# 5. Add to Claude Desktop config
# See QUICKSTART.md for claude_desktop_config.json setup

Installation

Prerequisites

• Node.js >=18.0.0
• At least one API key:
  • Anthropic Claude (recommended): ANTHROPIC_API_KEY
  • OpenAI GPT (alternative): OPENAI_API_KEY
  • Google Gemini (cost-effective): GOOGLE_API_KEY

Step-by-Step Installation

1. Clone Repository

git clone https://github.com/your-org/deep-research-mcp.git
cd deep-research-mcp

2. Install Dependencies

npm install

3. Configure Environment

cp .env.example .env

Edit .env and add your API key:

# Minimum required: ONE of these
ANTHROPIC_API_KEY=sk-ant-api03-...
# OPENAI_API_KEY=sk-...
# GOOGLE_API_KEY=...

# Optional: Configure defaults
DEFAULT_RESEARCH_MODE=standard
MAX_AGENTS=10

4. Build

npm run build

5. Verify Installation

npm run dev

Should output: [DeepResearch] ✓ Server ready

Claude Desktop Integration

Add to your claude_desktop_config.json:

Location:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Configuration:

{
  "mcpServers": {
    "deep-research": {
      "command": "node",
      "args": ["/absolute/path/to/deep-research-mcp/dist/index.js"],
      "env": {
        "ANTHROPIC_API_KEY": "sk-ant-api03-..."
      }
    }
  }
}

Restart Claude Desktop and verify the server appears in the MCP tools list.

Usage

Basic Research

Use deep-research MCP:
start_research(
  query: "What are the best practices for React hooks in 2025?",
  mode: "standard"
)

Output (after 10-15 minutes):

• Research ID for tracking
• Estimated cost (~$1.00)
• 4 agent reports (Industry, Docs, Community, Tools)
• 40+ sources with citations
• Confidence-scored findings

Research Modes

Advanced Usage

Custom Flags:

start_research(
  query: "Should we use microservices?",
  mode: "custom",
  flags: {
    agentCount: 6,
    ultrathink: true,
    minSources: 30,
    timeout: 30
  }
)

Check Status:

ping_research(researchId: "uuid-here")

Get Report:

get_report(researchId: "uuid-here", format: "markdown")

MCP Tools Reference

1. start_research

Start multi-agent research session.

Inputs:

• query (required): Research question
• mode (optional): quick | standard | deep | custom (default: standard)
• flags (optional): Configuration overrides

Returns:

• researchId: UUID for tracking
• estimatedTime: Minutes
• estimatedTokens: Token budget
• estimatedCost: USD

2. ping_research

Check status of active research.

Inputs:

• researchId (required): UUID from start_research

Returns:

• status: running | consolidating | completed | failed
• progress: 0-100%
• agentsCompleted: Count
• findingsCount: Total findings

3. get_report

Get full research report.

Inputs:

• researchId (required): UUID
• format (optional): markdown | json (default: markdown)

Returns:

• Executive summary
• Recommendation
• Agent reports (4-10)
• Quality metrics
• Citations
• Implementation steps

4. get_suggestions

Get proactive research suggestions.

Inputs:

• context (optional): Additional context
• limit (optional): Max suggestions (default: 5)

Returns:

• List of suggested research topics

Architecture

Multi-Agent System

Orchestrator
├─ Agent 1: Industry Best Practices
│  └─ Web search, GitHub trending, Academic papers
├─ Agent 2: Official Documentation
│  └─ Claude Code docs, Framework docs, APIs
├─ Agent 3: Community Insights
│  └─ Reddit, Discord, HackerNews, Twitter
├─ Agent 4: Tools & Repositories
│  └─ GitHub repos (>500 stars), npm, PyPI
└─ Consolidation Engine
   └─ Cross-reference validation → Unified report

Provider Abstraction

• Anthropic Claude: Best results, recommended
• OpenAI GPT: Alternative provider
• Google Gemini: Cost-effective option
• Fallback Chain: Automatic retry if provider fails

Key Design Principles

1. Fresh Context Per Agent: Prevents context rot
2. Evidence-Based Everything: Confidence scores + citations
3. Token Budget Control: Hard limits on spend
4. Parallel Execution: 4-10 agents simultaneously
5. Configurable Depth: User decides time/cost tradeoff

Configuration

Environment Variables

See .env.example for all available options.

Essential:

ANTHROPIC_API_KEY=sk-ant-...
DEFAULT_RESEARCH_MODE=standard
MAX_AGENTS=10

Advanced:

RESEARCH_TIMEOUT=900            # seconds
SOURCE_FILTER=all               # all | trusted | recent
RECENCY_FILTER=6months          # 6months | 1year | 2years
ENABLE_ULTRATHINK=true
MIN_SOURCES_PER_AGENT=10

Budget Control:

MAX_TOKENS_PER_SESSION=100000
MAX_COST_PER_SESSION=5.00      # USD

Mode Presets

Presets defined in src/config/modes.ts:

• QUICK_MODE: Fast validation
• STANDARD_MODE: Matches /deep-research behavior
• DEEP_MODE: Comprehensive analysis

Examples

Example 1: Technology Selection

Query: "Should I use LangChain, AutoGen, or CrewAI?"
Mode: standard
Result: Comparison table + recommendation + tradeoffs
Cost: ~$1.00

Example 2: Security Audit

Query: "Is express-session safe for production auth?"
Mode: quick
Result: Risk level + CVEs + mitigations
Cost: ~$0.35

Example 3: Migration Planning

Query: "How to migrate from Webpack to Vite?"
Mode: deep
Result: Step-by-step guide + gotchas + code examples
Cost: ~$3.00

Industry Gaps Addressed

Based on 2024-2025 research, this tool fills these gaps:

Troubleshooting

"No LLM providers initialized"

Cause: No API key configured Solution: Add at least one API key to .env:

ANTHROPIC_API_KEY=sk-ant-...

"Research session not found"

Cause: Invalid research ID or session expired Solution: Use researchId from start_research response

"Provider failed"

Cause: API key invalid or network issue Solution:

1. Check API key format
2. Verify network connectivity
3. Check provider status page

High Costs

Cause: Using deep mode frequently Solution:

1. Use quick or standard mode for most tasks
2. Set budget limits in .env:

   MAX_TOKENS_PER_SESSION=50000
   MAX_COST_PER_SESSION=1.00
   

Development

Build

npm run build

Watch Mode

npm run dev

Type Check

npm run type-check

Lint

npm run lint

Contributing

Contributions welcome! Please:

1. Fork the repository
2. Create feature branch (git checkout -b feature/amazing-feature)
3. Commit changes (git commit -m 'Add amazing feature')
4. Push to branch (git push origin feature/amazing-feature)
5. Open Pull Request

License

MIT License - see file

Support

• Issues: GitHub Issues
• Discussions: GitHub Discussions
• Documentation:

Roadmap

• Core multi-agent system
• Anthropic/OpenAI/Gemini support
• Configurable modes (quick/standard/deep)
• Security validation tool
• Comparison tool
• Persistent research cache
• Team knowledge sharing
• Web UI dashboard

Acknowledgments

• Inspired by the /deep-research slash command pattern
• Built on Model Context Protocol (MCP)
• Powered by LangChain

Made with ❤️ by the AI Lecture MCP Team