Wide research, not deep reports

Ground your AI agents in current reality.

12 sources in 40 seconds. Not one deep dive.

Daily coding is hundreds of micro-decisions: debug this error, check if that library is maintained, find the right API pattern. Deep research tools give you thorough reports—but you don't need a 10-minute analysis when you're mid-task.

Wide research synthesizes many sources fast. Think: asking 12 developers for quick takes vs. hiring one consultant for a report.

When to use each:

• Wide (95% of coding): Debugging, API lookups, library choices, error messages, compatibility checks
• Deep (5%): Learning new frameworks, architecture decisions, strategic planning

LLMs guess from outdated training data. When your agent needs current info—recent CVEs, breaking changes, deprecation notices—it's working from stale patterns instead of facts.

ArguSeek is an MCP server that gives AI agents real-time web research:

• 12+ sources in ~40 seconds — not one deep dive
• Bias detection — promotional content flagged with counter-queries
• Context chaining — build on previous queries without repeating work

Answers questions by researching multiple sources in parallel.

"Research passport.js JWT CVEs since January 2025" "What changed in React Server Components v14 to v15? Focus on breaking changes" 

How it works: Your query → 2-3 optimized Google searches → 30 URLs deduplicated → 12+ sources fetched → bias detection + synthesis → answer with citations.

Extracts specific information from a known URL (HTML or PDF).

fetch_url(url="https://docs.stripe.com/api/authentication", looking_for="authentication methods") fetch_url(url="https://github.com/vercel/next.js/releases/tag/v15.0.0", looking_for="breaking changes") 

# Install brew tap ArguSeek/arguseek && brew install arguseek # Or: git clone ... && make install # Configure export GOOGLE_API_KEY="your-key" export GOOGLE_CSE_ID="your-cse-id" # Run arguseek # stdio mode (for Claude Code, Claude Desktop)

claude mcp add arguseek arguseek

Done. Both tools are now available to your agent.

Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json):

{ "mcpServers": { "arguseek": { "command": "arguseek", "env": { "GOOGLE_API_KEY": "your-key", "GOOGLE_CSE_ID": "your-cse-id" } } } }

HTTP mode (containers, remote):

arguseek -http # Runs on :8080 claude mcp add --transport http arguseek http://localhost:8080/mcp

Query → Query Optimization (LLM) → 2-3 Parallel Google Searches → 30 URLs (deduplicated, ranked) → Fetch 12+ (two-phase fallback) → Bias Detection ‖ Synthesis (parallel) → Answer with citations 

• Deployment: DEPLOYMENT.md — Cloud Run, Docker, traffic verification
• Security: PRODUCTION_SECURITY.md — Auth strategies for production
• Development: CLAUDE.md — Full API reference, architecture details

MIT