spatial reasoning MCP tools

What the company appears to be: A self-reported tool built for agentic AI systems that provides deterministic spatial reasoning capabilities via a Model Context Protocol (MCP) server. The author describes it as an infrastructure layer enabling agents to compute real geometry instead of hallucinating spatial relationships.

What changed: This is a hackathon submission, not a product in production. It was built to address a gap identified in LLM-based agent behavior — specifically, that agents often hallucinate spatial relationships rather than computing them. The project aims to provide a standardized MCP interface for common geospatial operations like distance calculation, point-in-polygon testing, nearest neighbor search, and reachability analysis.

The single most important open question: Is there any evidence of actual adoption or integration by real AI agents beyond the author's own testing? The description states the tools were tested with Claude Desktop and Cursor but does not indicate whether these are live integrations or merely proof-of-concept demonstrations.

This is a self-reported, unverified account of a hackathon project. No revenue, customers, traction, or commercial evidence is provided.

The description states that the product is a Model Context Protocol (MCP) server exposing four deterministic geometry tools:

1. distance_between — computes real geodesic distance using the haversine formula.
2. is_within_zone — performs true point-in-polygon tests against named geographic zones.
3. nearest_of_type — finds k nearest points of interest ranked by real geodesic distance.
4. reachable_within — determines what's reachable on foot within N minutes using a cached street network.

It also includes a fifth tool, list_data_layers, which allows introspection of loaded data layers and auto-discovers features from GeoJSON files without code changes.

The implementation uses:

• Python
• Shapely for geometry
• NetworkX for graph algorithms
• OSMnx for walking network routing
• MCP Python SDK's FastMCP server

Inference: The product is described as a server-based infrastructure layer that can be called by any agent connected to the MCP protocol, similar to how Google Maps or Mapbox provide geospatial data.

The author claims this project addresses a gap in LLM-based agents' spatial reasoning capabilities — specifically, that they often hallucinate instead of compute geometry. The positioning is framed around:

• Grounded simplicity (like Ushahidi and Be My Eyes)
• Real geometry over pattern-matching
• Infrastructure-as-a-service for agents via MCP protocol

The project evolved from a hackathon idea to a tool designed to be integrated into existing AI workflows, not as an end-user application.

Inference: The author positions this as a low-level geospatial infrastructure that can be plugged into any agent system using the MCP protocol, rather than a standalone product or platform.

The description does not identify specific target customers or personas. It implies the primary users are:

• AI agents (specifically those using the MCP protocol)
• Developers building agentic systems
• Organizations deploying LLM-based agents in domains requiring spatial reasoning

Inference: The ICP is likely developers and teams working on agentic AI systems, particularly those needing reliable geospatial computation within their agent workflows.

No business model or pricing information is provided. The project is described as a hackathon submission with no indication of monetization strategy, licensing terms, or commercial use cases beyond the author’s own testing.

Not evidenced: No evidence of revenue streams, pricing tiers, subscription models, or commercial partnerships.

The description provides technical details:

• Built using Python, Shapely, NetworkX, OSMnx
• Exposed via MCP Python SDK's FastMCP server
• Uses real geospatial data (OpenStreetMap walking networks)
• Implements Dijkstra’s algorithm for reachability
• Supports auto-discovery of GeoJSON files

Challenges mentioned include:

• Serialization issues with GraphML and NetworkX
• Environment variable propagation across subprocesses
• Cache mutation bugs affecting tool outputs

Inference: The technical implementation shows a working prototype that addresses core geospatial computation needs, but lacks production-grade robustness or scalability indicators.

The description states:

• Tested with Claude Desktop and Cursor agents
• Unit-level sanity checks and standalone client scripts
• Live connection testing confirmed natural-language tool invocation

However, there is no evidence of:

• Real-world adoption
• Customer base
• Usage metrics
• Product maturity beyond prototype stage

Not evidenced: No traction or user engagement data.

The description does not mention competitors. It references two hackathon projects (Ushahidi and Be My Eyes) as examples of successful infrastructure built quickly, but does not describe existing tools in the geospatial or MCP space.

Inference: The competitive landscape is unclear. The author positions this as filling a gap in agent-based spatial reasoning, suggesting potential overlap with:

• Geospatial APIs (e.g., Google Maps, Mapbox)
• MCP protocol implementations
• LLM agent tooling platforms

1. Unproven adoption: No evidence of real-world usage or integration beyond author testing.
2. Prototype-only status: Described as a hackathon project with no indication of production readiness.
3. Limited scope: Only four core tools exposed; no mention of extensibility or broader functionality.
4. No commercialization path: No pricing, licensing, or monetization strategy described.
5. Technical complexity not fully validated: Bugs related to environment handling and serialization suggest potential issues in real-world deployment.

1. Has this tool been integrated into any live AI agent workflows beyond the author's own testing?
2. Are there plans for data ingestion from formats beyond GeoJSON (e.g., CSV, Shapefile)?
3. What is the current state of performance and scalability for large-scale use cases?
4. How does the project plan to handle updates to geospatial data sources like OpenStreetMap?
5. Is there any intention to expand beyond walking-based reachability into driving or multi-modal routing?