ddg-deep-research 0.2.0

5-stage deep research pipeline using DuckDuckGo MCP — free, no API key, no rate limits

🧠 ddg-deep-research

5-stage deep research pipeline using DuckDuckGo MCP — free, no API key, no rate limits.

pip install ddg-deep-research
ddg-deep-research search "your query"

Why this exists

Every other deep research agent requires OpenAI / Anthropic / Google API keys — or runs expensive local models. This one uses DuckDuckGo MCP for both search and content extraction. Completely free. Zero API keys. Zero rate limits.

Pipeline

Stage 1: Decompose  ──→ 3-6 sub-questions + search strategy
Stage 2: Gather     ──→ DuckDuckGo MCP search (parallel, unlimited)
Stage 3: Deep Read  ──→ DuckDuckGo MCP fetch_content + browser_use
Stage 4: Verify     ──→ Cross-reference claims, flag contradictions
Stage 5: Synthesize ──→ Cited brief.md + .provenance.md sidecar

Quick Start

# Search the web (free, no API key!)
ddg-deep-research search "latest advances in AI reasoning"

# Fetch a webpage
ddg-deep-research fetch "https://example.com/article"

Full Pipeline

# Stage 1: Break question into sub-questions (template)
ddg-deep-research decompose --question "How is RAG evolving?" --output plan.json

# Stage 2: Search
ddg-deep-research ddg_search --query "RAG architectures 2026" --output results.json

# Stage 3: Fetch content
ddg-deep-research ddg_fetch --url "https://..." --output-dir extracted/

# Stage 4: Merge results
ddg-deep-research merge --input-dir raw/ --output merged.json

# Stage 5: Clean & verify
ddg-deep-research clean --input extracted/ --output cleaned/
ddg-deep-research verify --input cleaned/cleaned.json --output verified.json

# Stage 6: Generate final brief
ddg-deep-research synthesize --verified verified.json --question "..." --output-dir outputs/ --today $(date +%Y-%m-%d)

Parallel DAG Execution

ddg-deep-research dag --plan workflow.json --verbose

Input JSON:

{
  "tasks": [
    {"id": "search_1", "depends_on": [], "command": "ddg_search --query ..."},
    {"id": "fetch_1", "depends_on": ["search_1"], "command": "ddg_fetch --url ..."},
    {"id": "synthesize", "depends_on": ["search_1", "fetch_1"], "command": "synthesize ..."}
  ]
}

Python API

import asyncio
from ddg_deep_research.ddg_mcp import search_web, fetch_content

async def main():
    results = await search_web("your query", max_results=10)
    for r in results:
        print(f"{r['title']}: {r['url']}")

    content = await fetch_content("https://example.com")
    print(content[:500])

asyncio.run(main())

Requirements

• Python 3.10+
• uv installed (for duckduckgo-mcp-server): curl -LsSf https://astral.sh/uv/install.sh | sh
• No API keys. No subscriptions. Nothing.

How it works

This package wraps duckduckgo-mcp-server via Python's MCP stdio transport. All search and content extraction goes through DuckDuckGo's free anonymous API. The 5-stage pipeline is modeled after production research agents but without the API costs.

Comparison

Feature	OpenAI Deep Research	LangChain Deep Research	ddg-deep-research
API key needed	✅ $200/mo	✅ OpenAI key	❌ Free
Search engine	Bing/Browser	Custom	DuckDuckGo
Content extraction	Built-in	Built-in	DuckDuckGo MCP
Provenance tracking	✅	✅	✅ .provenance.md
DAG orchestration	❌	❌	✅ Built-in
Open source	❌	✅	✅ MIT
pip install	❌	❌	✅ pip install

License

MIT