MCP Server · 8 tools · Free

Your AI understands
how things connect.

Konstruct gives Claude a knowledge graph that reasons about relationships between concepts. It doesn't just find information — it infers connections, triggers relevant frameworks, and generates structured guidance.

claude_desktop_config.json 30 sec setup

{
  "mcpServers": {
    "konstruct": {
      "url": "https://konstruct-api.aerynw.workers.dev/mcp"
    }
  }
}

What this looks like in practice

I'm building a marketplace app. How should I think about pricing strategy?

↻ konstruct_query "marketplace pricing strategy"

Concepts: marketplace_dynamics, pricing_strategy, network_effects, unit_economics
Edges: pricing_strategy → enables → growth, network_effects → amplifies → marketplace_dynamics
Framework triggered: YC Startup School — "marketplace" stage: growth
Advisory (287 tokens): "Marketplace pricing requires solving chicken-and-egg. Consider subsidizing supply side first..."

There's a structural relationship here worth understanding. Pricing strategy in a marketplace directly enables growth, but it's amplified by network effects. The YC framework flags this as a chicken-and-egg problem — you'll likely need to subsidize one side first. Let me walk you through the three common approaches and which fits your stage...

This isn't search. Konstruct didn't look up "marketplace pricing" — it walked typed edges between concepts, found that pricing_strategy enables growth which is amplified by network_effects, and triggered a YC framework because the matched concepts hit its activation conditions. The advisory was generated within a 300-token budget, ready for prompt injection.

Three ways teams use Konstruct

Structured knowledge that makes AI responses more precise and grounded.

Technical decision making

Load architecture knowledge packs. When you ask about database choices, Claude walks relationships — "PostgreSQL enables ACID which enables consistency" — and recommends based on your constraints.

konstruct_advisory "choosing database for fintech"

Framework: "Data Layer" → ACID required for financial data

Learning new domains

Explore unfamiliar territory by walking concept graphs. Start from one term, discover what it enables, requires, and contradicts — building mental models faster.

konstruct_explore "kubernetes" depth=2

→ enables: horizontal_scaling, service_discovery

Startup strategy guidance

Pre-loaded with YC, First Round, and a]16z frameworks. Ask about fundraising or go-to-market — get structured advice that triggers the right mental models for your stage.

konstruct_query "series A preparation"

Framework triggered: "Fundraising Readiness" stage: growth

8 tools for structured intelligence

Not a vector database. A relational inference engine that reasons about typed connections.

Query

konstruct_query — natural language graph query

konstruct_advisory — token-budgeted guidance

konstruct_explore — walk outward from concept

Knowledge

konstruct_load_pack — load a knowledge pack

konstruct_concept — get concept details

konstruct_edges — get typed relationships

19 typed relation types

Every edge has mathematical symmetry. Symmetric relations auto-create reverse edges. Invertible pairs generate inference.

causes prevents requires enables conflicts_with supports amplifies diminishes part_of is_a specializes generalizes produces consumes precedes follows analogous implements regulates

Symmetric Invertible pair Directional

The query pipeline

From natural language to structured advisory in microseconds.

Keyword extraction

Your query decomposes into concept identifiers via deterministic matching. No embeddings, no hallucinated connections. Consistent, reproducible, auditable.

Edge walking

From matched concepts, typed edges are traversed to configurable depth. Symmetry rules auto-generate reverse paths. Transitive chains surface connections you didn't ask about but should know.

Framework triggering

Matched concepts are checked against loaded frameworks — YC, Techstars, 500 Global, or custom. If triggered, structured guidance is generated within your token budget: questions, red flags, green lights.

Built in Rust. Compiled to WASM. Running on Cloudflare's edge. This isn't a wrapper around a vector database. Konstruct is a relational inference engine — queries use deterministic keyword matching against typed concept graphs. No embeddings, no hallucinated connections. Edge walking respects mathematical symmetry rules to generate consistent, auditable results. The MCP server runs a compiled WASM binary with zero cold starts. Knowledge is isolated per user via Durable Objects. This is the same code that powers Aerware Foundry's production knowledge layer.

Knowledge packs

Pre-built domain intelligence. Load what you need.

Startup Fundamentals

Core business concepts, market dynamics, product development, fundraising. YC, Techstars, 500 Global frameworks.

Ships free

Startup Accelerator

Growth patterns, unit economics, competitive analysis, pivot triggers. 40+ typed relationships.