agentshub

A collection of AI agents for trend research, news analysis, and content intelligence. Each agent can run standalone (as an MCP server) or together through an orchestrator that routes queries to the right agents automatically.

Agents

Example: site_security agent

> scan the security of example-shop.com

agents firing:
  → site_security(url='https://www.example-shop.com')
  [21:22:10] Scanning https://www.example-shop.com
  [21:22:11]   ✓ SSL valid, expires 2026-06-20 (68 days)
  [21:22:14] Checking security headers...
  [21:22:14]   0/7 headers present, 7 missing
  [21:22:14] Checking for WordPress...
  [21:22:16]   ⚠ EXPOSED: /wp-content/debug.log
  [21:22:20] Scan complete: 0 critical, 8 warnings, 1 info
  [21:22:20] Grade: C

Key findings:
  ✓ SSL valid (Let's Encrypt, 68 days remaining)
  ✓ HTTPS redirect working
  ✗ 7/7 security headers missing (HSTS, CSP, X-Frame-Options, etc.)
  ✗ Server header exposes software version
  ✗ /wp-content/debug.log publicly accessible

Install from PyPI

pip install agentshub-ai              # core agents
pip install agentshub-ai[instagram]   # + Instagram via Apify
pip install agentshub-ai[seo]         # + Google Search Console
pip install agentshub-ai[all]         # everything

# Register as MCP server in Claude Code
claude mcp add agentshub agentshub-mcp

Quick start (from source)

git clone https://github.com/ilarumk/agentshub.git
cd agentshub
python -m venv .venv && source .venv/bin/activate
pip install -e ".[all]"

# Add your keys
cp .env.example .env
# Edit .env with your GOOGLE_CLOUD_PROJECT, OPENAI_API_KEY, etc.

# Run the interactive chat (supervisor routes to agents automatically)
python chat.py

Three ways to use

1. Interactive chat — supervisor decides which agents to call

python chat.py

> what's trending in AI advertising this week?
# → fires rising_search + news_trending + social_trends automatically

2. Parallel demo — sequential vs parallel with timing

python demo.py
# → runs 3 agents sequential, then parallel, shows speedup

3. Standalone MCP — use one agent from Claude Code

# Register a single agent
claude mcp add rising-search python standalone/rising_search_mcp.py

# Or register all agents at once
claude mcp add agentshub python -m agentshub.mcp_server

Architecture

User query
    │
    ▼
Orchestrator / Supervisor (GPT-4o-mini)
    │ reads query → decides which agents to call
    │
    ├─→ rising_search     [BigQuery — no LLM]
    ├─→ news_trending     [RSS + web fetch — no LLM]
    ├─→ social_trends     [10 blogs + sub-agent GPT-4o-mini]
    ├─→ wikipedia_spike   [Pageview API — no LLM]
    ├─→ youtube_shorts    [YouTube API — no LLM]
    ├─→ instagram_trends  [Apify — no LLM]
    ├─→ search_console    [GSC API — no LLM]
    ├─→ bbc_news          [BigQuery — no LLM]
    ├─→ patent_search     [BigQuery — no LLM]
    ├─→ bigquery_analyst  [BigQuery — any dataset, any SQL]
    └─→ site_security    [HTTP checks — no auth needed]
    │
    ▼
Synthesizer (GPT-4o-mini) → final response

Adding a new agent

1. Create agentshub/agents/your_agent.py with a run(**kwargs) -> dict function
2. Add entry to REGISTRY in agentshub/agents/__init__.py
3. (Optional) Create standalone/your_agent_mcp.py for standalone use
4. The orchestrator, chat, and MCP server pick it up automatically

Stack

• Orchestration: Custom fan-out/fan-in with ThreadPoolExecutor
• Supervisor: GPT-4o-mini via OpenAI function calling
• Data agents: BigQuery, Google News RSS, Wikipedia API, YouTube API
• Sub-agent extraction: GPT-4o-mini for structured trend extraction
• MCP: Model Context Protocol for standalone agent access
• Web fetch: Parallel article fetching with blocking/paywall handling

Requirements

• Python 3.11+
• Google Cloud project with BigQuery access (for rising_search)
• OpenAI API key (for supervisor + social_trends sub-agent)
• YouTube API key (for youtube_shorts, optional)
• Apify API token (for instagram_trends, optional)
• Google Search Console access (for search_console, optional)