Deep Search

Free, Open-Source Search Engine MCP Server — 7 sources, 10 consolidated tools, semantic search via ChromaDB, zero cost.

Features

• 7 Data Sources: Web, Reddit, YouTube, GitHub, Twitter/X, DuckDuckGo, Wikipedia
• 10 Consolidated Tools: All features preserved, combined by mode/action parameters
• Semantic Search: ChromaDB + sentence-transformers (all-MiniLM-L6-v2)
• 100% Free: No API keys, no subscriptions, no paid APIs
• MCP Standard: Works with Claude, Cursor, OpenCode, and other AI clients
• Parallel Crawling: asyncio-based concurrent data collection

Installation

Option 1: Plugin Installation (Recommended)

OpenCode:

{
  "plugin": ["deep-search@git+https://github.com/sukirman1901/DeepSearch.git"]
}

Claude Code: Add to .mcp.json or ~/.claude/config.json:

{
  "mcpServers": {
    "deep-search": {
      "command": "python3",
      "args": ["server.py"],
      "cwd": "/path/to/DeepSearch/mcp"
    }
  }
}

Option 2: Manual Installation

# Clone the repository
git clone https://github.com/sukirman1901/DeepSearch.git
cd DeepSearch

# Create virtual environment
python3.12 -m venv .venv
source .venv/bin/activate

# Install dependencies
pip install -r mcp/requirements.txt

Available Tools (10)

search — Unified Search (7 modes)

Mode	Description	Key Params
basic (default)	Semantic search across indexed content	source, limit, category, search_depth, topic, max_age_hours
advanced	Search with domain/date/text/source filters	include_domains, exclude_domains, start_date, end_date
quick	Real-time search without database (DuckDuckGo)	source
stream	Search with streaming batches + timing	sources
smart	Compact IR overview + full details (saves 50-70% tokens)	top_full, max_overview_tokens
code	Search GitHub + Stack Overflow for code snippets	language, tokens_target
context	Token-budget-aware snippet packing	budget_tokens, language

crawl — Crawl & Extract

Mode	Description	Key Params
Single URL	Crawl URL + subpages, index results	url, subpages, subpage_target
Batch	Extract content from multiple URLs	urls, extract_depth, instructions

monitor — Persistent Monitoring

Action	Description
create	Create a monitor for a query
list	List all monitors
run	Run monitor, returns only NEW results
delete	Delete a monitor

webset — Entity Collection

Action	Description
create	Create a named container
add	Search and add results
list	List all websets
get	Get webset with all items
enrich	Scrape for emails, social links, tech
delete	Delete a webset

info — Engine Information

Type	Description
categories	List all search categories
sources	List all 7 data sources
stats	Database + cache statistics
detect	Auto-detect category for a query

research — Deep Research Sessions

Action	Description
start	Start a research session
followup	Ask follow-up question
list	List all sessions
delete	Delete a session

Other Tools

Tool	Description
answer	Search + synthesis with inline citations
search_leads	Lead generation with ICP scoring
site_map	BFS website structure mapping
index_topic	Crawl and index a topic

Architecture

DeepSearch/
├── mcp/                    # MCP server implementation
│   ├── crawlers/           # 7 specialized crawlers + subpage discovery
│   ├── db/                # ChromaDB + sentence-transformers
│   ├── search/            # Engine, answer, context, streaming, research, monitors, websets, sitemap, extract
│   ├── tests/             # 192 tests
│   ├── server.py          # 10 consolidated MCP tools
│   └── requirements.txt
├── skills/                # AI skills
│   └── using-deep-search/SKILL.md
├── hooks/                 # Session hooks
├── docs/superpowers/specs/ # Design specs
└── README.md

How It Works

1. Crawlers gather raw data from 7 sources (parallel async)
2. Sentence-transformers embeds text to 384-dim vectors
3. ChromaDB stores vectors in memory
4. Search engine performs semantic search
5. AI agent validates and summarizes results

AI Validates Results

Crawlers collect raw data. AI agent downstream validates, scores, and summarizes. Don't just trust crawler output.

Supported Platforms

• OpenCode - Plugin installation via plugin config
• Claude Code - MCP server configuration
• Cursor - Plugin installation
• Codex - Plugin installation
• Kimi Code - Plugin installation
• Gemini CLI - Extension support
• Any MCP-compatible client

License

MIT