Persistent, shared memory for AI coding agents.
Give Claude Code, Cursor, and MCP clients one searchable, auditable context layer for your team.

---

Quick start · Features · How it works · Documentation · MCP server

Why shared memory

AI coding assistants are useful, but project decisions and conventions often get trapped in separate chats. NexusMind is an open-source, self-hosted memory layer for AI agents: capture team knowledge once, then search and reuse it from compatible coding tools and projects.

Use it as persistent memory for Claude Code, Cursor, and MCP-compatible clients. Keep technical decisions, conventions, findings, and project context in one place, with an admin interface and an audit trail.

What you can do

• Share context across agents. Store and retrieve memories through the Model Context Protocol (MCP), so compatible assistants can work from the same team knowledge.
• Find relevant decisions. Search memories by project, tag, or content instead of repeating the same discovery in every conversation.
• See how knowledge connects. Explore related memories and projects in a knowledge graph.
• Keep teams in control. Manage users, roles, API keys, and audit events from the admin panel.
• Run it yourself. Start the backend and admin panel with Docker Compose; data is stored in SQLite by default.

Quick start

Start a local demo with Docker Compose:

make demo

Open http://localhost:3000 and sign in with the demo key printed by the command (nm_demo_acme_admin).

Heads-up: the demo setup rebuilds and reseeds apps/backend/data/nexusmind.db. Run it in a disposable local checkout, not against a database with data you need to keep.

For a manual development setup, environment requirements, and the MCP inspector, see Running NexusMind locally.

Connect an AI coding agent

Build the MCP server and set a demo member key:

make mcp-build
export NEXUSMIND_API_KEY=nm_demo_acme_sarah

Open Claude Code from the repository root. The included MCP configuration starts NexusMind with three tools:

Tool	Use it to
store_memory	Save a decision, convention, finding, or project note.
search_memory	Find relevant shared context.
list_memories	Browse recent memories and filter by project.

For example, ask your coding agent: “Remember that all REST endpoints use snake_case, and tag it as a convention.” Search for that decision later from another compatible agent.

How it works

flowchart LR
  Agent[Claude Code, Cursor, or MCP client] <-->|stdio| MCP[MCP server]
  MCP <-->|HTTP API and API key| API[Backend REST API]
  API <--> DB[(SQLite)]
  Admin[Admin panel] <-->|HTTP API| API

Loading

The MCP server connects coding agents to the backend. The admin panel uses the same API to manage team access, browse knowledge, and review activity.

Technology

Component	Technology
Backend API	Rust, Axum, SQLite
Admin panel	React, TypeScript, Vite
Agent integration	TypeScript, Model Context Protocol
Local deployment	Docker Compose

Development

# Backend
make backend

# Admin panel (in another terminal)
cd apps/admin && npm install && npm run dev

# MCP server
make mcp-build

# MCP smoke test (backend must be running)
NEXUSMIND_API_KEY=nm_demo_acme_sarah make mcp-test

See docs/RUNNING.md for prerequisites, ports, demo setup, and additional workflows.

Related projects

• nexusmind-mcp — standalone MCP server package for connecting coding agents to NexusMind.
• nexusmind-claude-plugin — Claude Code plugin with the MCP server, hooks, and skills.

License

MIT