N2NS Lab
Back to Blog
AI Collaboration•3 min read

Local-First AI Collaboration: Multi-Agent Coordination with n2n-nexus

How n2n-nexus coordinates multiple AI assistants and IDEs with local project context, tasks, and a shared message stream.

Written by N2NS Team
Published on 2026-06-20
Local-First AI Collaboration: Multi-Agent Coordination with n2n-nexus

Many developers do not use just one AI tool.

You might use Cursor to edit code, Claude Desktop to think through architecture, and Claude Code in your terminal to debug tests.

The problem is that these tools usually do not share state. Your Cursor agent may not know what your terminal assistant just fixed, which can lead to duplicate work, context drift, and messy git history.

n2n-nexus is a local-first coordination hub for that kind of multi-agent workflow.

The goal: One daemon. Many assistants. Shared project state. (Source: n2n-nexus README)

The Daemon-Adapter Architecture

To prevent agents from stepping on each other's toes, n2n-nexus splits tasks into two components: a central coordinator and stateless editors.

┌──────────────────────────────────────┐
│          n2n-nexus daemon            │
│  Standalone HTTP server · always on  │
│  Owns data, tools, tasks, messages   │
└──────────────┬───────────────────────┘
               │ HTTP (NEXUS_ENDPOINT)
       ┌───────┼───────┐
       ▼       ▼       ▼
     MCP-A   MCP-B   MCP-C
    (Win)   (WSL)   (SSH/VM)
   Stateless adapter per IDE

(Source: n2n-nexus README - Architecture)

  • The Daemon: A local HTTP server that runs in the background. It owns the database, records session history, manages shared documentation files, and tracks async tasks.
  • The Adapters: Lightweight, stateless MCP clients in your IDEs. They forward all agent actions to the daemon.

Because the system uses standard HTTP, it can handle cross-environment setups. Your Windows editor, WSL environment, and remote SSH servers can all point to the same daemon URL.

21 Tools for Collaboration

The daemon exposes 21 tools to keep your agents aligned:

Category Purpose
Session & Topology Declares your active project and maps dependencies.
Projects & Assets Manages configuration files, metadata, and assets.
Global Docs Syncs architecture blueprints and strategy files.
Messaging & Meetings Logs discussions, decisions, and meeting transcripts.
Tasks Spawns and monitors long-running background tasks.
Maintenance Cleans up system logs and deletes legacy projects.

(Source: n2n-nexus README - Toolset)

Everything is stored locally on your machine (under ~/.n2n-nexus/ by default). The system uses prefix-based project IDs—like web_main or api_auth—to keep things organized.

Multi-Agent Collaboration in Action

Here is what a coordinated workflow looks like:

  1. Context Alignment: Each AI assistant registers its scope with register_session_context to lock in read/write permissions.
  2. Structured Communication: Instead of posting raw text, agents flag their messages:
    • PROPOSAL for architecture suggestions.
    • DECISION to document finished plans.
    • UPDATE for build notifications.
  3. Cross-IDE Debugging: When your front-end agent hits an API error, it posts an update. The backend agent in your terminal sees the message, fixes the config, and replies. They can work through issues asynchronously without sharing one long chat transcript.
  4. Activity Logs: The daemon records events as [Augment] entries, tracking file syncs and meeting resolutions automatically.

(Source: n2n-nexus README - Real-world example)

Quick Start

Start the daemon on your machine:

npx n2n-nexus daemon --port 5688

Then configure your MCP client to connect to it:

{
  "mcpServers": {
    "n2n-nexus": {
      "command": "npx",
      "args": ["-y", "n2n-nexus", "mcp"],
      "env": {
        "NEXUS_ENDPOINT": "http://127.0.0.1:5688"
      }
    }
  }
}

(Source: n2n-nexus README - Quick Start)

Keep Shared State Local

When several assistants touch the same project, they need a shared place for context, tasks, and decisions. n2n-nexus keeps that coordination local instead of pushing every working note into a hosted service.

Visit the n2n-nexus GitHub repository to view the full docs.