sramkrishna/muninn-mcp-server

Muninn - Memory MCP Server

Muninn (Old Norse: Memory) is an MCP (Model Context Protocol) server that provides persistent memory and semantic search capabilities for agentic AI systems.

Named after Odin's raven of memory (paired with Hugin, the raven of thought), Muninn stores and recalls contextual information to enable intelligent desktop automation and workflow learning.

Features

• Hybrid Storage: SQLite for structured queries + ChromaDB for semantic search
• Local-First: All data and embeddings stay on your machine
• Privacy-Focused: Uses local embedding models (no cloud API calls)
• MCP-Compatible: Works with any MCP client (Hugin, Claude Desktop, etc.)

Architecture

┌─────────────────────────────────────────┐
│            Muninn Memory Server          │
├─────────────────────────────────────────┤
│                                          │
│  ┌──────────────┐    ┌───────────────┐ │
│  │   SQLite     │    │   ChromaDB    │ │
│  │  (Structured)│    │   (Semantic)  │ │
│  └──────────────┘    └───────────────┘ │
│         ↑                    ↑          │
│         └────────┬───────────┘          │
│                  │                      │
│         ┌────────▼────────┐            │
│         │  Local Embedder │            │
│         │ sentence-trans. │            │
│         └─────────────────┘            │
│                                          │
│  Storage: ~/.local/share/muninn/        │
└─────────────────────────────────────────┘

Installation

cd muninn-mcp-server
python3.13 -m venv .venv
source .venv/bin/activate
pip install -e .

Usage

Standalone

muninn

With Hugin (MCP Client)

Add to config.toml:

[servers.muninn]
command = "python3.13"
args = ["/path/to/muninn-mcp-server/src/muninn_mcp_server/server.py"]

Available Tools

Storage

• store_event - Store desktop events with automatic embedding
• store_pattern - Store detected behavioral patterns
• store_decision - Log agent decisions and outcomes

Retrieval (Structured)

• query_events - SQL-based queries on structured data
• get_recent_events - Get last N events
• get_events_by_type - Filter by event type

Retrieval (Semantic)

• semantic_search - Find similar events by meaning
• get_context - Get relevant context for current situation

Analytics

• detect_patterns - Find recurring patterns in stored data
• get_statistics - Aggregated statistics

Data Storage

All data stored locally at:

~/.local/share/muninn/
├── muninn.db           # SQLite database
└── chroma/             # Vector embeddings

Example: Store Desktop Event

# Via MCP client (e.g., Hugin + LLM)
await call_tool("store_event", {
    "event_type": "extension_change",
    "data": {
        "extension": "dash-to-panel",
        "action": "disabled",
        "reason": "causing crashes"
    },
    "description": "Disabled dash-to-panel extension due to repeated GNOME Shell crashes"
})

Example: Semantic Search

# Find similar past issues
await call_tool("semantic_search", {
    "query": "extension stability problems",
    "limit": 5
})

Privacy

• No cloud APIs: All embeddings generated locally
• No telemetry: All data stays on your machine
• No external dependencies: SQLite and ChromaDB run locally

Embedding Model

Uses all-MiniLM-L6-v2 (80MB) by default:

• Fast on CPU
• Good quality embeddings
• Runs entirely offline

License

MIT