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

Graph tools

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:

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

Skills (Multi-Step Workflows)

Skills chain multiple tools into one call, returning ready-to-use intelligence packages:

Why Forage over other MCP search tools?

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.

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