Forage-mcp
MCP server giving AI agents real-time web search, page scraping, company intelligence, email discovery, local lead generation, and a persistent knowledge graph. Pay only for what you use, no subscriptions.
README
Forage MCP — Web Intelligence & Persistent Knowledge Graph for AI Agents
By Riccardo Minniti / Ernesta Labs | riccardo@ernestalabs.com
Forage is a Model Context Protocol (MCP) server that gives AI agents real-time web intelligence and a self-accumulating knowledge graph. One connection provides 24 tools and 12 multi-step skills: web search, company data, verified B2B emails, local leads, and a graph that remembers everything your agent has ever discovered.
Built on Apify's scraping infrastructure. Powered by FalkorDB for persistent graph storage.
Author: Riccardo Minniti
Organization: Ernesta Labs
Contact: riccardo@ernestalabs.com
GitHub: github.com/ErnestaLabs/web-intelligence-mcp
Apify: apify.com/ernesta_labs/forage
The Knowledge Graph — Your Agent's Memory
Every tool call automatically feeds a private knowledge graph that grows smarter over time. No other MCP server does this.
┌─────────────────────────────────────────────────────────────┐
│ FORAGE KNOWLEDGE GRAPH │
├─────────────────────────────────────────────────────────────┤
│ │
│ [Company A] ──── has_domain ────▶ [domain.com] │
│ │ │ │
│ │ works_at │ has_email_pattern │
│ ▼ ▼ │
│ [Person B] ◀─── has_title ──── [Email Pattern] │
│ │ │ │
│ │ located_in │ verified_emails │
│ ▼ ▼ │
│ [San Francisco] [john@domain.com] │
│ │ │ │
│ │ operates_in │ linkedin │
│ ▼ ▼ │
│ [SaaS Industry] [LinkedIn Profile] │
│ │
│ Claims: "Raised Series A in 2024" (confidence: 89%) │
│ Signals: Hiring spike in Q4 2025 (+45%) │
│ Regime: growth │
│ │
└─────────────────────────────────────────────────────────────┘
What makes it different
| Feature | Forage Graph | Generic RAG/Vector DB |
|---|---|---|
| Entity deduplication | ✓ SHA-256 identity | ✗ Creates duplicates |
| Relationship tracking | ✓ Typed edges (works_at, located_in) | ✗ Similarity only |
| Confidence scoring | ✓ Increases with corroboration | ✗ Static embeddings |
| Provenance (claims) | ✓ Who said what, when | ✗ No source tracking |
| Time-series signals | ✓ Track metrics over time | ✗ Snapshot only |
| Causal inference | ✓ Find what drives what | ✗ No causality |
| Regime detection | ✓ Normal / stressed / pre-tipping | ✗ No state tracking |
Graph tools
| Tool | What it does | Price |
|---|---|---|
query_knowledge |
Search entities by name/type | $0.05 |
enrich_entity |
Full profile + all relationships | $0.08 |
find_connections |
Path between two entities | $0.12 |
add_claim |
Store provenance assertion | $0.05 |
get_claims |
Retrieve claims for entity | $0.05 |
add_signal |
Record time-series data point | $0.05 |
get_signals |
Query metrics over time | $0.05 |
set_regime |
Label entity state | $0.03 |
get_regime |
Check entity state | $0.03 |
causal_parents |
What drives this entity | $0.08 |
causal_children |
What this entity drives | $0.08 |
causal_path |
Highest-weight causal path | $0.15 |
simulate |
Propagate shock/boost through graph | $0.25 |
get_graph_stats |
Entity/relationship counts | Free |
The graph is persistent — stored in FalkorDB on our infrastructure. Your agent's research accumulates across sessions. The more you use Forage, the smarter it gets.
Email Verification — How It Actually Works
We don't just guess email patterns. Each find_emails call runs a 4-step verification pipeline:
Step 1: Pattern Discovery
Scrape the target domain for email patterns (e.g., firstname.lastname@domain.com). Extract from:
- Contact pages, footers, team pages
- Press releases, blog author pages
- WHOIS records, SSL certificates
Step 2: Candidate Generation
Generate candidate emails using discovered patterns + LinkedIn data. Cross-reference with:
- Company employee listings (if public)
- Job postings with contact info
- Conference speaker lists
Step 3: SMTP Verification
For each candidate, we perform an SMTP handshake check:
- Connect to the domain's mail server
- Verify the recipient exists (
RCPT TO) - Detect catch-all domains (score penalty)
- Detect mailboxes that accept then bounce (honeypots)
Step 4: Confidence Scoring
Each email gets a confidence score (0-100) based on:
| Signal | Weight | Example |
|---|---|---|
| SMTP accept | 40% | Mail server accepted RCPT TO |
| Pattern match | 25% | Matches known company format |
| LinkedIn match | 20% | Name matches LinkedIn profile |
| Source corroboration | 15% | Found on multiple public sources |
Return format:
{
"email": "sarah.chen@stripe.com",
"name": "Sarah Chen",
"title": "VP of Sales",
"seniority": "vp",
"department": "sales",
"linkedin": "linkedin.com/in/sarahchen",
"confidence": 94,
"verified": true,
"verification_steps": ["smtp_accepted", "linkedin_match", "pattern_match"]
}
What "verified" means
- Confidence 90-100: SMTP accepted + LinkedIn match + multiple sources. High deliverability.
- Confidence 70-89: SMTP accepted or strong pattern match. Good for outreach.
- Confidence 50-69: Pattern-based with partial verification. Use with caution.
- Below 50: Not returned (filtered out).
This is not a simple mailserver check. It's a multi-source corroboration pipeline that other MCP servers don't offer.
Web Intelligence Tools
Core Tools
| Tool | What it does | Price | Why this price |
|---|---|---|---|
search_web |
Multi-source search, deduplicated, ranked | $0.03 | Aggregates Brave, Bing, DuckDuckGo + dedup + rank. Cheaper than calling each API separately ($0.03 total vs $0.06+ if you called 2 search APIs) |
scrape_page |
Extract clean markdown from any URL | $0.07 | Includes proxy rotation, JavaScript rendering, anti-bot bypass |
get_company_info |
Domain → full company profile | $0.08 | Aggregates 5+ data sources: website, LinkedIn, Crunchbase patterns, social profiles |
find_emails |
Verified B2B emails with LinkedIn | $0.10 | 4-step pipeline above |
find_local_leads |
Local businesses by niche + location | $0.15 | Google Maps + enrichment + phone/website extraction |
find_leads |
B2B leads by title/industry/location | $0.25/100 leads | That's $0.0025 per lead. Try finding 100 leads manually. |
Skills (Multi-Step Workflows)
Skills chain multiple tools into one call, returning ready-to-use intelligence packages:
| Skill | Price | Returns |
|---|---|---|
skill_company_dossier |
$0.50 | Full company profile + 10 contacts with emails |
skill_prospect_company |
$0.75 | 15 decision makers sorted by seniority + emails |
skill_outbound_list |
$3.50 | 100 verified leads ready for CRM import |
skill_local_market_map |
$0.80 | Up to 60 local businesses with contact info |
skill_decision_maker_finder |
$1.00 | 20 decision makers by seniority tier |
skill_competitor_intel |
$0.80 | Pricing, features, reviews, positioning |
skill_competitor_ads |
$0.65 | Active ad copy, landing pages, platforms |
skill_job_signals |
$0.55 | Hiring trends, open roles, expansion signals |
skill_tech_stack |
$0.45 | Technologies used with confidence scores |
skill_funding_intel |
$0.70 | Funding rounds, investors, valuation estimates |
skill_social_proof |
$0.55 | Reviews, ratings, testimonials aggregated |
skill_market_map |
$1.20 | Complete competitor landscape for a market |
Why Forage over other MCP search tools?
| Capability | Forage | Brave Search MCP | Apify MCP | AgentQL |
|---|---|---|---|---|
| Web search | ✓ | ✓ | ✗ | ✗ |
| Page scraping | ✓ | ✗ | ✓ | ✓ |
| Email discovery | ✓ (4-step verified) | ✗ | ✗ | ✗ |
| B2B leads | ✓ | ✗ | Partial | ✗ |
| Company intelligence | ✓ | ✗ | Partial | ✗ |
| Local businesses | ✓ | ✗ | ✓ | ✗ |
| Persistent knowledge graph | ✓ | ✗ | ✗ | ✗ |
| Provenance & claims | ✓ | ✗ | ✗ | ✗ |
| Causal analysis | ✓ | ✗ | ✗ | ✗ |
| Time-series signals | ✓ | ✗ | ✗ | ✗ |
| Multi-step skills | ✓ (12 skills) | ✗ | ✗ | ✗ |
| Actor gateway (1000+) | ✓ | ✗ | ✓ | ✗ |
The knowledge graph is the differentiator. Other tools give you data. Forage gives you accumulated intelligence. Every search, every email lookup, every company profile feeds your private graph. After a week of use, your agent knows more about your market than any single search ever could.
Quick Start
1. Get Your API Token
Go to Apify Console → Settings → Integrations and copy your Personal API Token.
2. Connect to Claude / Cursor / n8n
Claude Desktop (%APPDATA%\Claude\claude_desktop_config.json):
{
"mcpServers": {
"forage": {
"command": "npx",
"args": [
"-y", "@anthropic/mcp-proxy",
"https://ernesta-labs--forage.apify.actor/mcp/sse"
],
"env": {
"APIFY_API_TOKEN": "YOUR_APIFY_TOKEN"
}
}
}
}
Cursor / Windsurf:
{
"mcpServers": {
"forage": {
"command": "npx",
"args": [
"-y", "@anthropic/mcp-proxy",
"https://ernesta-labs--forage.apify.actor/mcp/sse"
],
"env": {
"APIFY_API_TOKEN": "YOUR_APIFY_TOKEN"
}
}
}
}
n8n / LangGraph / Custom: Connect to the SSE endpoint at https://ernesta-labs--forage.apify.actor/mcp/sse with your Apify token in the Authorization header.
3. System Prompt (Optional)
Add to your agent's system prompt:
When you need live web data, company info, verified emails, or lead lists, use Forage tools. Each call costs money (shown in responses), so batch operations when possible. Your knowledge graph persists across sessions — check it first before making new web calls.
Examples
Find 20 HVAC leads in Dallas
Call: find_leads
Params: { "industry": "HVAC", "location": "Dallas, TX", "limit": 20 }
Returns: 20 companies with name, phone, website, email, address
Cost: $0.05
Get decision makers at a prospect
Call: skill_prospect_company
Params: { "domain": "stripe.com" }
Returns: 15 decision makers with title, email, LinkedIn, seniority
Cost: $0.75
Build a local market map
Call: skill_local_market_map
Params: { "business_type": "dentist", "location": "London, UK" }
Returns: 60 dentists with address, phone, website, rating, reviews
Cost: $0.80
Track a company's hiring over time
Call: add_signal
Params: { "entity": "Acme Corp", "metric": "job_postings", "value": 45 }
... repeat weekly ...
Call: get_signals
Params: { "entity": "Acme Corp", "metric": "job_postings" }
Returns: Time-series of job postings — hiring trend visible
Find who influenced a deal
Call: find_connections
Params: { "from": "Your Company", "to": "Acme Corp" }
Returns: Path through shared connections, events, technologies
Architecture
┌─────────────────────────────────────────────────────────────┐
│ YOUR AI AGENT (Claude, Cursor, n8n) │
└─────────────────────────┬───────────────────────────────────┘
│ MCP Protocol
▼
┌─────────────────────────────────────────────────────────────┐
│ FORAGE MCP SERVER (Apify) │
│ ┌────────────┐ ┌────────────┐ ┌──────────────────────┐ │
│ │ Web Search │ │ Scraper │ │ Email Discovery │ │
│ │ (3 engines)│ │ (rendered) │ │ (4-step pipeline) │ │
│ └─────┬──────┘ └─────┬──────┘ └──────────┬───────────┘ │
│ └───────────────┼────────────────────┘ │
│ ▼ │
│ ┌─────────────────┐ │
│ │ Graph Client │ │
│ └────────┬────────┘ │
└───────────────────────┼────────────────────────────────────┘
│ HTTPS
▼
┌─────────────────────────────────────────────────────────────┐
│ FORAGE GRAPH API (Railway) │
│ ┌──────────────────────────────────────────────────────┐ │
│ │ FalkorDB (Redis-compatible) │ │
│ │ Entities ──── RELATES ────▶ Entities │ │
│ │ Claims (provenance) │ │
│ │ Signals (time-series) │ │
│ │ Regimes (state tracking) │ │
│ └──────────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────┘
Pricing
Pay per tool call. No subscription. No minimum. Every response includes the cost.
| Your Apify spend | Your Forage cost | Ratio |
|---|---|---|
| $1 | ~$0.75 | 25% markup |
| $10 | ~$7.50 | 25% markup |
| $100 | ~$75 | 25% markup |
The 25% markup covers: proxy infrastructure, knowledge graph storage, email verification pipeline, multi-engine search aggregation, and ongoing maintenance.
Free trial: New Apify accounts get $5 platform credit. Try Forage risk-free.
Limitations
- Some sites block scraping — we use proxies + JS rendering, but some sites (LinkedIn, closed social networks) are protected
- Email accuracy ≠ 100% — confidence scores reflect real verification, but email addresses can change
- Knowledge graph is persistent but not portable — data lives on our FalkorDB instance (not exported yet)
- Rate limits — Apify enforces per-account limits; Forage doesn't add extra limits on top
Support & Links
- Author: Riccardo Minniti / Ernesta Labs
- Email: riccardo@ernestalabs.com
- GitHub: github.com/ErnestaLabs/web-intelligence-mcp
- Issues: github.com/ErnestaLabs/web-intelligence-mcp/issues
- Apify Actor: apify.com/ernesta_labs/forage
- Documentation: See QUICKSTART.md and EXAMPLES.md
License
MIT License — Copyright (c) 2026 Riccardo Minniti / Ernesta Labs
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。