One shared brain for every AI on your team — persistent across sessions
Mono Memory MCP
One shared brain for every AI on your team — persistent across sessions, searchable, always in sync.
A lightweight, self-hosted MCP server that gives your AI coding assistants long-term memory. Built for teams where multiple people use AI-powered editors (Claude Code, Cursor, Windsurf) and need their AIs to remember past decisions, share discoveries, and stay aligned — without re-explaining everything every session.
The Problem
- Your AI assistant forgets everything when a session ends.
- Each team member's AI works in isolation — no shared knowledge.
- Critical decisions, bug fixes, and architectural context get lost between sessions.
The Solution
Mono Memory gives your team's AI assistants a shared, persistent memory backed by a single SQLite file. Any AI can save and retrieve observations, project context, and decisions — across sessions, across team members.
Why "Mono"? — Like a monorepo manages all code in one place, Mono Memory manages all your team's AI knowledge in one server.
How It Works
Session 1 (Alice — morning)
├─ AI discovers a tricky bug in auth logic
├─ → memory_save: "JWT refresh token race condition fix — added mutex lock"
└─ Session ends. AI forgets everything.
Session 2 (Bob — afternoon)
├─ AI starts working on auth-related feature
├─ → memory_search: "auth"
├─ ← Gets Alice's bug fix context instantly
└─ Avoids the same pitfall, builds on her solution.
Session 3 (Alice — next day)
├─ → memory_timeline: project="my-app", since="2025-03-01"
└─ ← Sees everything the team's AIs learned this week.
Every observation is stored in a shared SQLite database. Any team member's AI can save and query it through 6 MCP tools.
Use Cases
Solo Developer
- Session continuity — Your AI remembers yesterday's debugging insights, architectural decisions, and TODO notes without you copy-pasting context.
- Project context — Store your project's architecture, conventions, and API specs once. Your AI loads them on demand instead of re-reading files every session.
Team (2-10 developers)
- Shared knowledge base — One person's AI discovers a gotcha? Everyone's AI knows about it.
- Onboarding — New team members' AIs instantly access the full history of decisions and patterns.
- Cross-project awareness — Working on the frontend? Search what the backend team's AI learned about the API yesterday.
Multi-project
- Centralized memory — One server, multiple projects. Search across all or filter by project.
- Timeline view — See the evolution of decisions across your entire organization.
Features
- 6 tools — save, get, search, timeline, init, context
- SQLite storage — zero-config, WAL mode, single-file database
- Streamable HTTP — network-ready transport for team use
- Environment variable config — host, port, database path
- Multi-project — multiple authors and projects, keyword search, timeline view
Quick Start
There are two roles: Host (runs the server) and Client (connects via plugin).
Host: Start the Server
The host is the person (or machine) that runs the Mono Memory server for the team.
git clone https://github.com/potato-castle/mono-memory-mcp.git
cd mono-memory-mcp
uv run python server.py
The server starts on http://0.0.0.0:8765/mcp (streamable-http). Share this URL with your team — replace 0.0.0.0 with your machine's IP address (e.g. http://192.168.0.10:8765/mcp).
Custom configuration:
# Change port
MONO_MEMORY_PORT=9000 python server.py
# Change database directory
MONO_MEMORY_DB_DIR=/path/to/data python server.py
# Run in background
nohup python server.py > /tmp/mono-memory.log 2>&1 &
Client: Install the Plugin (Claude Code)
Clients do not need to clone the repo. Just run three commands in Claude Code:
1. Register the marketplace:
/plugin marketplace add potato-castle/mono-memory-mcp
2. Install the plugin:
/plugin install mono-memory-mcp@mono-memory-mcp
When prompted for scope, select "Install for you, in this repo only (local scope)". This keeps the plugin active only in the current project.
3. Run the setup skill:
/mono-memory-mcp:setup
This will prompt you for:
- Server URL — the host's server address (e.g.
http://192.168.0.10:8765/mcp) - Author name — your name, used to tag memories you save
The project name is automatically detected from your current directory name.
The setup will:
- Write
.mcp.jsonin your project root (MCP server connection) - Append auto-recording rules to
CLAUDE.md(so your AI automatically saves discoveries)
Restart Claude Code to activate.
Tools
`memory_save` — Save an observation
Store a discovery, decision, debugging insight, or any knowledge.
| Parameter | Required | Description |
|---|
Tools (6)
memory_saveStore a discovery, decision, debugging insight, or any knowledge.memory_getRetrieve specific memory entries.memory_searchSearch through stored memories using keywords.memory_timelineView the evolution of decisions and discoveries over time.memory_initInitialize the memory database for a project.memory_contextRetrieve relevant context for the current project.Environment Variables
MONO_MEMORY_PORTThe port the server runs onMONO_MEMORY_DB_DIRThe directory path for the SQLite databaseConfiguration
{"mcpServers": {"mono-memory": {"command": "python", "args": ["/path/to/mono-memory-mcp/server.py"]}}}