Idea Lab

An MCP server that turns your AI assistant into a structured ideation partner. It generates, scores, critiques, and stores software ideas locally — so your assistant remembers past ideas across sessions and never suggests the same thing twice.

Includes a built-in web dashboard with a Kanban board, lineage graph, and portfolio view — with dark and light mode support.

What it does

You talk to your AI assistant like normal. Idea Lab gives it tools to:

Generate ideas using 4 different creative techniques, then run each one through a pipeline:

"Generate 5 software ideas in the developer tools space"

  idea_lab_generate_ideas  →  generates candidates
  idea_lab_save_idea       →  stores each one
  idea_lab_score_idea      →  scores on 7 dimensions
  idea_lab_critique_idea   →  tries to tear it apart
  idea_lab_check_duplicate →  checks against everything stored

Search and revisit ideas you've generated before:

"Show me my highest-rated ideas"
"Find ideas related to automation"
"What shortlisted ideas haven't I looked at in a while?"

Evolve ideas that survived the pipeline:

"Take that CLI tool idea and reimagine it as a SaaS product"
"Create an MVP plan for my top-rated idea"

Everything is stored locally in SQLite. No cloud, no API keys, no data leaving your machine.

Related MCP server: Doclea MCP

Web Dashboard

The MCP server includes a built-in web dashboard that starts automatically at http://localhost:3001. No separate process needed.

• Kanban Board — drag-and-drop ideas between status columns

• Lineage Graph — visualize how ideas connect through mutations

• Portfolio View — heat map and table showing domain distribution and scores

• Detail Panel — click any idea for its full breakdown (radar chart, critique flags, MVP steps, lineage)

• Dark / Light Mode — toggle in the navbar, respects OS preference, persists across sessions

Use the idea_lab_open_dashboard tool to open it in your browser, or just navigate to http://localhost:3001 directly.

For the full guide, see Web Dashboard Guide.

Install

From npm

npm install idea-lab-mcp
cd node_modules/idea-lab-mcp
node build/db/migrate.js

From source

git clone https://github.com/mordiaky/idea-lab-mcp.git
cd idea-lab-mcp
npm install
npm run build        # compiles server + web dashboard
npm run db:migrate

Requires Node.js 22+. If npm install fails, see platform setup.

Once connected to your editor, the MCP server starts automatically and the web dashboard is available at http://localhost:3001.

Connect to your editor

Add the server to your MCP client. Replace /absolute/path/to with the real path to where you cloned the project.

Create .vscode/mcp.json:

{
  "servers": {
    "idea-lab": {
      "type": "stdio",
      "command": "node",
      "args": ["/absolute/path/to/idea-lab-mcp/build/index.js"]
    }
  }
}

Create ~/.cursor/mcp.json (global) or .cursor/mcp.json (per-project):

{
  "mcpServers": {
    "idea-lab": {
      "command": "node",
      "args": ["/absolute/path/to/idea-lab-mcp/build/index.js"]
    }
  }
}

Edit ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "idea-lab": {
      "command": "node",
      "args": ["/absolute/path/to/idea-lab-mcp/build/index.js"]
    }
  }
}

Add to ~/.continue/config.yaml:

mcpServers:
  - name: idea-lab
    command: node
    args:
      - /absolute/path/to/idea-lab-mcp/build/index.js

Use the MCP Servers panel in Cline's UI, or edit the config directly:

{
  "mcpServers": {
    "idea-lab": {
      "command": "node",
      "args": ["/absolute/path/to/idea-lab-mcp/build/index.js"]
    }
  }
}

{
  "mcpServers": {
    "idea-lab": {
      "command": "node",
      "args": ["/absolute/path/to/idea-lab-mcp/build/index.js"]
    }
  }
}

Fully quit and restart the app after editing.

claude mcp add idea-lab -- node /absolute/path/to/idea-lab-mcp/build/index.js

Add to ~/.config/zed/settings.json:

{
  "context_servers": {
    "idea-lab": {
      "source": "custom",
      "command": "node",
      "args": ["/absolute/path/to/idea-lab-mcp/build/index.js"]
    }
  }
}

Example workflows

Brainstorm session

"Generate software ideas for the healthcare space, then score and critique the best ones"

The assistant will:

1. Call generate_ideas to create 5-8 candidates using different creative techniques

2. Save each one and score it across 7 dimensions (novelty, usefulness, feasibility, testability, speed to MVP, defensibility, clarity)

3. Critique the high scorers — searching for existing products, fragile dependencies, and vague claims

4. Check for duplicates against your stored ideas

5. Return a summary of what passed and what got rejected

Build on a winner

"Show me my shortlisted ideas and create an MVP plan for the best one"

The assistant will:

1. Search your stored ideas filtered to shortlisted status

2. Pick the highest-scoring one (or let you choose)

3. Generate 3-5 concrete MVP steps with tech stack recommendations

Explore a different angle

"Take idea X and mutate it — what if we targeted enterprise users instead?"

The assistant calls mutate_idea to create a variant that changes the target user while keeping the core concept. The variant is linked to the original so you can compare lineage.

Import ideas from your notes

"Here are my ideas from last month's brainstorm session: [paste messy notes]"

The assistant calls submit_idea with your raw text, extracts each discrete idea, structures them, and runs each through the full scoring pipeline.

Strengthen a weak idea

"Idea X scored weak — refine it with a 'weekend-buildable' constraint"

The assistant calls refine_idea to apply the constraint, generating a tighter version that gets re-scored. You can keep applying different constraints (CLI-only, offline-only, etc.) until it passes or you move on.

Break a big idea into shippable pieces

"Decompose my top idea into things I can actually start building"

The assistant calls decompose_idea to split it into 3-7 independently shippable micro-products, each scored on its own. It recommends which piece to start with this weekend.

Revisit old ideas

"What shortlisted ideas have I not looked at in 2 weeks?"

The assistant calls resurface_ideas to find stale shortlisted ideas that might deserve a second look now that your thinking has evolved.

How scoring works

Each idea is scored 0-10 on 7 dimensions with different weights:

Ideas need to clear minimum thresholds (feasibility >= 7, usefulness >= 7, novelty >= 6, composite >= 6.5) to advance.

How ideas move through the pipeline

raw  -->  shortlisted  -->  build-next
 |             |                |
 +---- rejected (terminal) ----+

• raw — just generated, not yet evaluated

• shortlisted — passed scoring, critique, and dedup checks

• build-next — picked for MVP planning

• rejected — didn't survive evaluation (can happen at any stage)

Tools reference

Configuration

Both settings are optional. The defaults work without any configuration.

Development

npm run dev           # Run MCP server (no build step, serves built dashboard)
npm run build         # Compile TypeScript server + React dashboard
npm test              # Run tests
npm run db:generate   # Generate migrations from schema changes
npm run db:migrate    # Apply migrations
npm run db:seed       # Load sample data

For dashboard frontend development with hot reload:

cd web && npm run dev  # Vite dev server, proxies API to port 3001

Part of Thinking Tools

Idea Lab is also bundled in Thinking Tools MCP — a unified suite with 61 tools across 8 modules (including Hypothesis Tracker, Decision Matrix, Mental Models, and more). Use this standalone version if you only need ideation.

License

PolyForm Noncommercial 1.0.0 — free for personal and non-commercial use.

For commercial licensing, contact the author.