igenius-mcp 0.5.1

MCP server for iGenius Memory — gives AI agents persistent memory tools via the hosted API

iGenius MCP — Thin Memory Server for VS Code

A lightweight MCP (Model Context Protocol) server that gives AI agents persistent memory via the iGenius Memory service.

Core memory tools require only an API key — all AI processing happens server-side. Optional visual tools add local Playwright rendering and vision-model analysis.

Quick Start

pip install igenius-mcp

Set your API key and run:

export IGENIUS_API_KEY=ig_your_key_here
igenius-mcp

Get a free key at igenius-memory.online

VS Code Setup

Add to ~/.vscode/mcp.json:

{
  "servers": {
    "igenius-memory": {
      "command": "igenius-mcp",
      "env": { "IGENIUS_API_KEY": "ig_your_key_here" },
      "type": "stdio"
    }
  }
}

Restart VS Code — all 17 memory tools are now available to Copilot and any MCP-compatible agent.

Available Tools

Tool	Description
memory_briefing	Session briefing from all memory layers (call FIRST)
memory_ingest	Ingest user/agent messages for AI extraction
memory_consolidate	Merge accumulated extracts into master briefing
memory_process	Detect trigger words and auto-classify text
memory_store	Direct store to a specific memory layer
memory_search	Natural language search across memories
memory_recall	Retrieve all persistent session memories
memory_summarize	LLM-powered summary of a memory layer
memory_delete	Delete a memory by ID
memory_update	Update fields on an existing memory
memory_review	List short-term memories for triage
memory_promote	Promote short-term → long-term
memory_pin	Pin a fact permanently (user-confirmed, never expires)
memory_triggers_list	List trigger words and their layers
memory_triggers_add	Add a new trigger word

LLM Requirements

iGenius uses an LLM backend for AI extraction, consolidation, and (optionally) visual analysis. You can use a local or remote LLM provider.

Local Setup (LM Studio, Ollama, etc.)

Requirement	Minimum
GPU VRAM	6 GB+
Recommended model	Qwen 3.5 4B (non-thinking) or equivalent
Context window	3,000+ tokens

⚠️ IMPORTANT: Do NOT use thinking/reasoning models (e.g. QwQ, DeepSeek R1, o1, o3). Thinking models emit <think> chains before the actual response, which breaks iGenius's structured JSON extraction pipeline. Only use standard non-thinking (instruct/chat) models.

Why these specs? iGenius sends structured extraction prompts that expect clean JSON output. A 4B-parameter non-thinking model at 3k context is the sweet spot for fast, accurate extraction without hallucination or timeouts. Larger models (8B+) work too — just ensure you have the VRAM headroom and that the model is a non-thinking variant.

Remote Setup (OpenAI, Anthropic, Google, etc.)

No local hardware requirements. Any API-accessible model works — configure the provider, model name, and API key in the VS Code extension settings or environment variables.

Environment Variables

Variable	Required	Default
IGENIUS_API_KEY	Yes	—
IGENIUS_API_URL	No	https://igenius-memory.online/v1

Visual Tools (Optional)

Give your AI agent eyes — render any URL, take a pixel-perfect screenshot, and get instant UI/UX analysis from a local vision model.

Install

pip install "igenius-mcp[visual]"
python -m playwright install chromium

Then load a vision-capable model in LM Studio (e.g. Qwen 3.5 9B Vision, non-thinking).

⚠️ Do NOT use thinking/reasoning vision models — same restriction as above.

Visual MCP Tools

Tool	Description
visual_report	Render URL → screenshot → vision analysis → full UI/UX report
visual_screenshot	Render URL → return base64-encoded PNG (no analysis)

Visual Environment Variables

Variable	Default	Description
IGENIUS_VISION_URL	http://localhost:1234/v1	Vision model API endpoint
IGENIUS_VISION_MODEL	auto-detect	Override the vision model name
IGENIUS_VIEWPORT_W	1280	Screenshot viewport width
IGENIUS_VIEWPORT_H	800	Screenshot viewport height

100% local — screenshots and analysis never leave your machine.

Agent Instructions

For best results, add the iGenius agent instructions to your workspace:

• VS Code: Place igenius.instructions.md in ~/.vscode/prompts/
• Claude Code: Add to CLAUDE.md
• Workspace: Add to .github/copilot-instructions.md

Get the template at igenius-memory.info

How It Works

Agent ←→ MCP (stdio) ←→ igenius-mcp ←→ REST API ←→ iGenius Backend

The memory tools are a thin proxy — they translate MCP tool calls into REST API requests. All AI extraction, LLM summarization, and encryption happens server-side.

The visual tools run locally — Playwright renders URLs on your machine and a local vision model (e.g. LM Studio + Qwen2.5-VL) analyzes the screenshots. Screenshots and analysis never leave your machine.

Plans

Plan	Price	Requests	API Keys	IPs/Key
Starter	Free	1,000/week	1	3
Pro	$19/mo	50,000/day	5	10
Enterprise	Contact	500,000/day	20	50

Details at igenius-memory.store

Links

• Landing Page
• Documentation
• API Portal
• Store & Plans

Support the Project

iGenius Memory is built and maintained by NovaMind Labs. If you find it useful, here's how you can help:

• Star the repo — it helps more developers discover iGenius
• Upgrade to Pro — $19/mo directly funds development → igenius-memory.store
• Report bugs & ideas — open an issue
• Spread the word — tell your friends, tweet about it, write a blog post

Every user, star, and subscription helps keep iGenius alive and improving. Thank you!

License

MIT