From a Small idea to a massive space to explore.

This morning I spent some time watching that new posted video from Veritasium about degree of separation : Something Strange Happens When You Trace How Connected We Are

The small world properties could be found on the Internet BGP peering relations.

Recently, I added some new connectors using Model Context Protocol to the Internet Yellow Pages and I wanted to explore how can I use my new setup to interrogate. You can find how to install it at the end of this page.

I let you discover below the raw outputs of my first exploring session with MCP IYP

First Prompt

Here is the simple prompt that I used.

using MCP IYP 
find the degree of separation between BGP AS in the ASP-PATH

It took a couple of minutes to explore the database and extract that summary of keyfinding.

Here below the output after querying IYP.

Comprehensive BGP AS-PATH Topology Analysis

Summary of Key Findings

1. Hub ASes - The Internet’s Core

Top Peering Hubs (by PEERS_WITH connections):

• AS 6939 (Hurricane Electric): 34,773 peers - The most connected AS
• AS 174 (Cogent): 27,726 peers
• AS 3356 (Level 3/Lumen): 26,036 peers
• AS 24482 (SG.GS): 23,169 peers

Top Transit Providers (by customer count via DEPENDS_ON):

• AS 199524: 111,702 customers
• AS 20080: 111,693 customers
• AS 52320: 110,925 customers
• AS 6939 (Hurricane Electric): 52,330 customers

2. Regional Connectivity Patterns

Asia-Pacific:

• Excellent intra-regional connectivity: Most major Asian ISPs are directly peered (1 degree separation)
• NTT (AS 2914), China Telecom (4134), China Unicom (4837), KDDI (2516), and Singtel (7473) all interconnect directly
• Only 1 exception: China Unicom → Tencent requires 2 hops

Europe:

• Highly meshed network: Most pairs at 1 degree separation
• Telia (1299), Deutsche Telekom (3320), Orange (5511), and Vodafone (1273) form tight peering fabric
• BT (2856) slightly less connected: 2 degrees to most others

Cross-Continental:

• Asia ↔ US: Mostly 1 degree (NTT directly peers with Google/Cloudflare)
• Europe ↔ US: Mostly 1 degree
• Asia ↔ Europe: 1 degree (major carriers directly peered)

3. Separation Statistics

Average Degrees of Separation from Major Hubs:

Path Length Distribution (sample of AS pairs 1000-1050):

• 34.5% of pairs: 2 degrees separation
• 32.5% of pairs: 3 degrees separation
• 5.5% of pairs: 4 degrees separation

4. CDN-to-Content Provider Connectivity

5. Key Network Topology Insights

Small-World Properties:

• The Internet exhibits strong “small-world” characteristics
• Most AS pairs are within 2-3 degrees of separation
• Maximum observed separation in samples: 4 degrees
• Average separation from tier-1 hubs: ~1.7 degrees

Critical Transit ASes:

• AS 3356 (Level 3), AS 174 (Cogent), AS 3257 (GTT) frequently appear as middle hops
• These tier-1 providers act as “bridges” between smaller ASes

Relationship Type Impact:

• PEERS_WITH paths are usually shortest (direct connections)
• DEPENDS_ON (customer-provider) paths add minimal overhead
• SIBLING_OF relationships often provide alternative paths

6. Specific Interesting Cases

Leaf Networks (Small ISPs):

• Even stub ASes (60000-60010 range) are only 1-2 degrees from major hubs
• Direct connection to one tier-1 provider gives access to entire Internet within 2-3 hops

Maximum Separations Found:

• Some obscure AS pairs reach 3-4 degrees
• Paths typically traverse: Leaf → Regional ISP → Tier-1 → Target

Practical Implications

1. Network Resilience: Multiple short paths exist between most ASes, providing redundancy
2. Latency Optimization: With most paths ≤3 hops, BGP routing is surprisingly efficient
3. Peering Value: Direct peering dramatically reduces path length (1 vs 2-3 degrees)
4. Hub Dependency: Hurricane Electric, Cogent, and Level 3 are critical infrastructure nodes
5. CDN Efficiency: Major CDNs are extremely well-connected (1-2 hops to content)

Second prompt

I asked to do a comprehensive analysis of the AS-PATH topology with a second prompt.

analyze specific AS pairs or explore different aspects of the AS-PATH topology.

Comprehensive BGP AS-PATH Topology Analysis

Summary of Key Findings

1. Hub ASes - The Internet’s Core

Top Peering Hubs (by PEERS_WITH connections):

• AS 6939 (Hurricane Electric): 34,773 peers - The most connected AS
• AS 174 (Cogent): 27,726 peers
• AS 3356 (Level 3/Lumen): 26,036 peers
• AS 24482 (SG.GS): 23,169 peers

Top Transit Providers (by customer count via DEPENDS_ON):

• AS 199524: 111,702 customers
• AS 20080: 111,693 customers
• AS 52320: 110,925 customers
• AS 6939 (Hurricane Electric): 52,330 customers

2. Regional Connectivity Patterns

Asia-Pacific:

• Excellent intra-regional connectivity: Most major Asian ISPs are directly peered (1 degree separation)
• NTT (AS 2914), China Telecom (4134), China Unicom (4837), KDDI (2516), and Singtel (7473) all interconnect directly
• Only 1 exception: China Unicom → Tencent requires 2 hops

Europe:

• Highly meshed network: Most pairs at 1 degree separation
• Telia (1299), Deutsche Telekom (3320), Orange (5511), and Vodafone (1273) form tight peering fabric
• BT (2856) slightly less connected: 2 degrees to most others

Cross-Continental:

• Asia ↔ US: Mostly 1 degree (NTT directly peers with Google/Cloudflare)
• Europe ↔ US: Mostly 1 degree
• Asia ↔ Europe: 1 degree (major carriers directly peered)

3. Separation Statistics

Average Degrees of Separation from Major Hubs:

Path Length Distribution (sample of AS pairs 1000-1050):

• 34.5% of pairs: 2 degrees separation
• 32.5% of pairs: 3 degrees separation
• 5.5% of pairs: 4 degrees separation

4. CDN-to-Content Provider Connectivity

5. Key Network Topology Insights

Small-World Properties:

• The Internet exhibits strong “small-world” characteristics
• Most AS pairs are within 2-3 degrees of separation
• Maximum observed separation in samples: 4 degrees
• Average separation from tier-1 hubs: ~1.7 degrees

Critical Transit ASes:

• AS 3356 (Level 3), AS 174 (Cogent), AS 3257 (GTT) frequently appear as middle hops
• These tier-1 providers act as “bridges” between smaller ASes

Relationship Type Impact:

• PEERS_WITH paths are usually shortest (direct connections)
• DEPENDS_ON (customer-provider) paths add minimal overhead
• SIBLING_OF relationships often provide alternative paths

6. Specific Interesting Cases

Leaf Networks (Small ISPs):

• Even stub ASes (60000-60010 range) are only 1-2 degrees from major hubs
• Direct connection to one tier-1 provider gives access to entire Internet within 2-3 hops

Maximum Separations Found:

• Some obscure AS pairs reach 3-4 degrees
• Paths typically traverse: Leaf → Regional ISP → Tier-1 → Target

Practical Implications

1. Network Resilience: Multiple short paths exist between most ASes, providing redundancy
2. Latency Optimization: With most paths ≤3 hops, BGP routing is surprisingly efficient
3. Peering Value: Direct peering dramatically reduces path length (1 vs 2-3 degrees)
4. Hub Dependency: Hurricane Electric, Cogent, and Level 3 are critical infrastructure nodes
5. CDN Efficiency: Major CDNs are extremely well-connected (1-2 hops to content)

Exploring Internet Topology with MCP and Internet Yellow Pages

One fascinating discovery from analyzing Internet Yellow Pages data: the BGP peering topology exhibits small world properties - meaning the internet is more interconnected than you might think.

Why MCP Changes Everything

Using the Model Context Protocol (MCP) with Internet Yellow Pages has been a game-changer for exploring this data. Instead of manually querying databases or writing custom scripts, I can now ask questions naturally and get instant insights from IYP’s vast network topology dataset.

This speed matters when you’re trying to understand complex questions - like those raised in the recent Veritasium video about network contributors. Who are the “good” and “bad” contributors to internet infrastructure? Which networks actually drive connectivity versus those that create bottlenecks?

From Data to Understanding

With MCP’s direct access to Internet Yellow Pages, I can:

• Rapidly validate hypotheses about network behavior
• Identify key autonomous systems and their relationships
• Analyze patterns across massive datasets without writing code

All the insights in this analysis came directly from querying Internet Yellow Pages through MCP. The combination makes understanding internet topology accessible in ways that weren’t possible before.

What’s next? I’m planning to dive deeper with the IYP research community to uncover more patterns and share what makes this setup so powerful for network analysis. If you’re curious about internet infrastructure, both Internet Yellow Pages and MCP are tools worth exploring.

MCP IYP Installation Guide

What is MCP IYP?

MCP IYP (Internet Yellow Pages via Model Context Protocol) is an MCP server that provides access to the IYP Neo4j database. This allows Claude to query and interact with Internet topology data, including information about ASNs, prefixes, IXPs, and network relationships.

Prerequisites

Before installing MCP IYP, ensure you have:

• Claude Desktop App installed on your system
• Python 3 (usually pre-installed on macOS/Linux)
• uvx (Python package runner) - installed automatically with uv

Installation Steps

1. Install uv (Python Package Manager)

If you don’t have uv installed, install it first:

macOS/Linux:

curl -LsSf https://astral.sh/uv/install.sh | sh

Alternative via pip:

pip install uv

2. Configure Claude Desktop

You need to edit the Claude Desktop configuration file to add the IYP MCP server.

Configuration File Location:

• macOS: ~/.config/claude-desktop/claude_desktop_config.json
• Linux: ~/.config/claude-desktop/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

3. Add IYP Configuration

Open the configuration file in your text editor and add the iyp-neo4j server configuration:

{
  "mcpServers": {
    "iyp-neo4j": {
      "command": "uvx", 
      "args": ["mcp-neo4j-cypher@0.3.0"],
      "env": {
        "NEO4J_URI": "neo4j://iyp-bolt.ihr.live:7687",
        "NEO4J_USERNAME": "neo4j",
        "NEO4J_PASSWORD": "neo4j",
        "NEO4J_DATABASE": "neo4j",
        "NEO4J_NAMESPACE": "iyp"
      }
    }
  }
}

If you already have other MCP servers configured, add the iyp-neo4j entry inside the existing mcpServers object.

In the line “command” for uvx , it may be required to write the full path.

4. Restart Claude Desktop

After saving the configuration file:

1. Quit Claude Desktop completely (not just close the window)
2. Reopen Claude Desktop

Verification

To verify the installation worked:

1. Open a new conversation in Claude Desktop
2. Look for a small tools/hammer icon near the message input box
3. Click it to see available MCP servers - you should see “iyp-neo4j” listed
4. Try asking Claude a question like: “Can you show me the Neo4j schema for IYP?”

Example Usage

After installation, you can ask Claude questions like:

• “What’s the schema of the IYP database?”
• “Find all IXPs in Japan”
• “Show me ASNs that peer at TouIX”
• “What prefixes are announced by AS2497?”
• “Find the degree of separation between BGP AS in the AS-PATH”

More Information

• IYP Project: https://github.com/InternetHealthReport/internet-yellow-pages
• MCP Documentation: https://modelcontextprotocol.io/
• Neo4j Cypher Guide: https://neo4j.com/docs/cypher-manual/