muninn-mcp 0.1.0

Cross-tool project memory for solo builders. Odin's raven remembers.

Muninn

"Odin feared losing Muninn more than Huginn — losing memory is worse than losing thought."

Cross-tool project memory for AI assistants. Muninn (MOO-nin) is an MCP server that gives Claude, ChatGPT, Cursor, and other AI tools shared, persistent, project-structured memory.

The Problem

Every tool switch = context reset. Every session = re-explanation from scratch.

• 2 hours brainstorming in Claude Desktop → switch to Claude Code → blank slate
• Research in ChatGPT → switch to Cursor → copy-paste everything
• Come back tomorrow → "this project is X, tech stack is Y, we're at Z..." again

How Muninn Solves It

Save in Claude Desktop    →  "Save this to ouri-app"
Switch to Claude Code     →  muninn_recall("ouri-app") → full context loaded
Research in ChatGPT       →  "Save findings to ouri-app"
Come back next week       →  Everything remembered

Zero copy-paste. Zero re-explanation. Works across any MCP-compatible tool.

Quick Start

Install

# From source (recommended during alpha)
git clone https://github.com/ilwonyoon/muninn.git
cd muninn
pip install -e .

Connect to Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "muninn": {
      "command": "/path/to/muninn/.venv/bin/python",
      "args": ["-m", "muninn.server"]
    }
  }
}

Restart Claude Desktop. Start saving memories.

Connect to Claude Code

claude mcp add muninn -- /path/to/muninn/.venv/bin/python -m muninn.server

Remote Access (HTTP Transport)

For Claude Web/Mobile, ChatGPT, or remote access:

# Basic (no auth — local dev only)
muninn --transport http --port 8000

# With authentication (recommended)
export MUNINN_API_KEY="your-secret-key"
muninn --transport http --host 0.0.0.0 --port 8000

Endpoint: http://localhost:8000/mcp

For public access, expose via ngrok or Cloudflare Tunnel:

ngrok http 8000
# Then add the HTTPS URL to claude.ai or ChatGPT as a remote MCP server

Tools

Tool	Purpose
muninn_save	Save memory to a project
muninn_recall	Load project context with depth filtering
muninn_search	Full-text search across all memories
muninn_status	List all projects with overview
muninn_manage	Delete/update memories, manage projects

Depth System

Muninn uses a 4-level depth system that works for any project type:

Depth	Question	Use For
0	"What is this?"	10-second project overview. Always loaded. Create first.
1	"To continue"	Resume work next session. Key decisions, current state.
2	"To go deeper"	Detailed analysis, specs, research. Loaded on request.
3	"Just in case"	Archive, raw data, logs. Rarely loaded.

muninn_recall(depth=0)  →  Project summaries only (~500 chars)
muninn_recall(depth=1)  →  Above + working context (default)
muninn_recall(depth=2)  →  Above + deep details
muninn_recall(depth=3)  →  Everything

Example Usage

In any AI tool connected to Muninn:

User: "Save this to ouri-app: voice journaling app for Korean millennials,
       React Native + Whisper API, freemium model"

→ muninn_save(project="ouri-app", content="...", depth=0)
→ Saved to ouri-app (memory: a1b2c3d4)

User: "What are my projects?"

→ muninn_status()
→ ouri-app    active  today       memories: 9
   muninn     active  today       memories: 15
   glpuri     paused  2 weeks ago memories: 23

User: "Load ouri-app context"

→ muninn_recall(project="ouri-app")
→ ## ouri-app (active)
  - [summary] (a1b2c3d4) Voice journaling app for Korean millennials...
  - [context] (b2c3d4e5) Tech stack: React Native + Whisper API...
  - [context] (c3d4e5f6) Revenue model: freemium with 5min/day limit...
  Context: 2,341 / 8,000 chars | 9 memories loaded

Data Storage

Default: ~/.local/share/muninn/muninn.db
Override: MUNINN_DB_PATH=/path/to/custom.db

SQLite with WAL mode. Single file. Back up with cp muninn.db muninn.db.bak.

Development

# Install with dev deps
pip install -e ".[dev]"

# Run tests (104 tests)
python -m pytest tests/ -v

# Run the server (stdio)
python -m muninn.server

# Run the server (HTTP)
python -m muninn.server --transport http --port 8000

Architecture

src/muninn/
  server.py      # MCP entry point (stdio + HTTP)
  auth.py        # Bearer token auth middleware
  tools.py       # 5 MCP tool functions
  store.py       # SQLite with FTS5 full-text search
  models.py      # Frozen dataclasses
  formatter.py   # LLM-readable output formatting

Platform Support

Platform	Transport	Status
Claude Desktop (Mac)	stdio	Working
Claude Code	stdio	Working
Cursor	stdio	Working
Claude Web/Mobile	HTTP	Ready (needs hosting)
ChatGPT Web	HTTP	Ready (needs hosting)
ChatGPT Mac	—	No MCP support yet

License

MIT