hypertopos-mcp

MCP tools for exploring geometric data spaces with AI agents.

hypertopos-mcp gives AI agents a way to explore data built with hypertopos. Instead of writing queries, agents navigate a geometric space — finding anomalies, tracing relationships, comparing populations, and tracking change over time.

Quick start

pip install hypertopos-mcp

Configure your MCP client:

{
  "mcpServers": {
    "hypertopos": {
      "command": "hypertopos-mcp",
      "env": {
        "HYPERTOPOS_SPHERE_PATH": "path/to/your/sphere"
      }
    }
  }
}

The agent starts with open_sphere, then either:

• detect_pattern("find anomalous accounts") — smart detection in a single call
• sphere_overview() — unlock the full manual toolset for step-by-step exploration

What an agent can do

• Find anomalies — entities far from the population norm
• Discover clusters and structural archetypes
• Navigate between related entities
• Trace relationship chains and transaction flows
• Compare populations across groups or time windows
• Track drift and detect regime changes
• Score contagion risk through network proximity

How it works

1. Build a geometric space from relational data using hypertopos-py
2. Connect the space via hypertopos-mcp
3. The agent explores — each finding leads to the next

Tools are registered dynamically. The agent starts with a small set and unlocks more as it explores — keeping context lean.

Works best with skills

hypertopos-mcp provides tools. hypertopos-skills provides judgment — structured investigation workflows that guide agents through real tasks like fraud detection, anomaly triage, and drift monitoring.

Documentation

Tool Reference	All tool parameters, return shapes, filters
MCP Specification	Server spec, lifecycle, transport, error codes

Status

Research-stage project. Tooling and API may evolve.

License

Apache License 2.0