Novelty EngineGet started

Find what others miss

Discover the gaps.Build what’s next.

Novelty Engine uses research, signals, and reasoning to uncover real market gaps and generate opportunities worth building.

Install skillLearn more
Scroll to explore

01 Novelty without demand is a curiosity.

02 Demand without differentiation is a crowded category.

03 Evidence without falsification is just a story.

About

Novelty Engine
Research System

A system for discovering real product-market gaps
through structured, falsifiable research.

Structured Research

Define a gap, collect evidence, and build a falsifiable case.

Falsifiable Evidence

Every claim is tested against verifiable, real-world data.

Clear Outcomes

From insight to opportunity with execution-ready clarity.

01 Novelty without demand is a curiosity.

02 Demand without differentiation is a crowded category.

03 Evidence without falsification is just a story.

Install

Install once. Connect the
research path that fits.

Install the Claude Skill locally, or connect the deployed /api/mcp
endpoint for live research without local configuration.

Option 01

Claude Skill

Install the local Skill.

Run both commands beside the Novelty Engine folder to add it to Claude’s local skills directory.

Manual install
mkdir -p ~/.claude/skills
cp -R novelty-engine ~/.claude/skills/
Option 02

Claude browser

Connect live research once.

Install the Skill, then open Settings → Connectors → Add custom connector and paste the deployed MCP endpoint. No Tavily key or local environment variable is needed.

Streamable HTTP endpoint
https://novelty-engine.vercel.app/api/mcp

Test connectionOpen health check

  1. Install the Skill

    Copy the folder into Claude’s local skills directory.

  2. Connect research

    Add the MCP URL in Claude browser, or set the direct helper URL in Claude Code.

  3. Ask normally

    Invoke /novelty-engine or request market-gap research.

The evidence loop

A research method, not a longer prompt.

The engine makes the search process inspectable from first source to final experiment.

  1. 01Map

    Competitors, complaints, workarounds, shifts

  2. 02Find

    Structural holes and underserved workflows

  3. 03Challenge

    Defaults, assumptions, and constraints

  4. 04Generate

    Candidates with a traceable lineage

  5. 05Falsify

    Demand, economics, trust, and feasibility

  6. 06Return

    Survivors with a measurable first test

A worked example

From a common prompt to a specific opening.

V2.1 records how an idea was derived, what resembles it, what could kill it, and the cheapest test that should come next.

Prompt

“Find opportunities for small field-service teams that still move job data between tools.”

First-pass brainstorming

Familiar answers

  1. 01
    All-in-one CRM

    Replace every current tool with one larger system.

  2. 02
    AI dispatcher

    Add a chat assistant on top of the existing workflow.

  3. 03
    Scheduling dashboard

    Show another view of data workers already maintain.

The subject changed. The underlying mechanisms did not.

With Novelty Engine

Exception Bridge

Survivor

A local-first event bridge keeps existing systems authoritative and asks a worker to intervene only when their records disagree—without introducing another CRM.

Evidence lineage
Duplicate entry → spreadsheets and messages → small-team pricing mismatch → remove the central database → local event capability.
Decisive risk
It fails if connectors are expensive to maintain, exceptions are too rare, or teams will not grant source-system access.
Validate first
Manually bridge ten real jobs for three teams in 72 hours. Continue only if errors fall by half and one team commits to a paid pilot.

Illustrative, fixture-backed pipeline walkthrough—not a claim about a live market.

The method

Research. Structure. Contradict. Falsify.

Compact provenance and visible tradeoffs replace vague claims of uniqueness.

  1. 01

    Build the landscape

    Map competitors, products, complaints, workarounds, technologies, and rules into an opportunity graph.

    Look for repeated friction, sparse combinations, underserved segments, failed attempts, and early change signals.

  2. 02

    Contradict the category

    Extract the assumptions a market treats as fixed, then systematically invert or remove the consequential ones.

    Transfer mechanisms from distant domains without simply borrowing their language or aesthetics.

  3. 03

    Generate with lineage

    Create a broad candidate field from evidence-backed holes, then reject ideas that collapse into familiar products.

    Every candidate keeps a compact record of the evidence, contradiction, and mutation that produced it.

  4. 04

    Try to kill the ideas

    Pressure-test demand, economics, distribution, feasibility, behavior, trust, regulation, and defensibility.

    Promising failures get at most two bounded mutations. Survivors leave with a 24–72 hour validation test.

Built for honest research

No competitor found does not mean a good opportunity was found.

When web access is configured, the backend searches competitors, reviews, forums, GitHub, directories, regulations, failed attempts, and counterevidence.

Missing evidence stays unknown. Novel proposals are hypotheses until a real market test says otherwise.

Novelty Engine V2.1

Make the next idea earn its place.

Install locally View on GitHub