Getting Started

Nexus is an enterprise dev toolkit with three capabilities: a Design-to-Code pipeline, 22 Software Development Lifecycle (SDLC) dev-workflow agents, and multi-step workflow commands — all powered by Claude Code.

Overview

Nexus ships as both a CLI (nexus) and an MCP server. Both share the same underlying tools.

Three tooling areas:

The MCP server additionally exposes nexus_search and nexus_read — hybrid web search and doc-optimised reading tools for AI agents (MCP only, not available in the CLI).

Prerequisites

# macOS (via Homebrew)
$ brew install python

# or download from python.org

$ curl -LsSf https://astral.sh/uv/install.sh | sh

# Install Claude Code
$ npm install -g @anthropic-ai/claude-code

# Authenticate
$ claude login

# Verify
$ claude --version

Installation & Integration

Pick the use case that matches what you want to do — each one is self-contained.

I want to generate code from a Figma Make export or prompt

Install the nexus CLI and run the Design-to-Code pipeline from your terminal.

$ curl -fsSL https://nexus.coderstudio.co/install.sh | bashcopy

$ npm install -g @anthropic-ai/claude-code
$ claude login

# From a text prompt
$ nexus run prompt "A SaaS dashboard" -g nextjs-fullstack -p my-app

# From a Figma Make ZIP
$ nexus run zip ~/Downloads/figma-export.zip -g nextjs-fullstack -p my-appcopy

I want to use Nexus tools inside a Claude Code session

Register Nexus as an MCP server — no separate install needed, uvx handles it on demand.

$ claude mcp add nexus -- uvx --refresh --from nexus-toolkit@latest nexus-mcpcopy

$ claude mcp list   # nexus ✓ Connected

I want to use Nexus in Claude Desktop or Cursor

Add one entry to your app's config file — no separate install needed.

{
  "mcpServers": {
    "nexus": {
      "command": "uvx",
      "args": ["--refresh", "--from", "nexus-toolkit", "nexus-mcp"]
    }
  }
}copy

I want to automate workflows with n8n

Run Nexus as a persistent HTTP MCP server — your n8n instance calls its tools to automate code review, security scanning, QA gates, and more.

$ MCP_TRANSPORT=http MCP_HOST=0.0.0.0 MCP_PORT=3900 uvx --from nexus-toolkit nexus-mcpcopy

http://<server-ip>:3900/mcp

Updating

Auto-update (recommended for uvx setups)

Add --refresh to your uvx command once — Nexus will check PyPI for a new version on every startup and update automatically if one is available. No manual steps after a release.

Claude Code: If you already added the server, remove it first then re-add with @latest:

$ claude mcp remove nexus
$ claude mcp add nexus -- uvx --refresh --from nexus-toolkit@latest nexus-mcpcopy

Claude Desktop / Cursor — edit your config file:

{
  "mcpServers": {
    "nexus": {
      "command": "uvx",
      "args": ["--refresh", "--from", "nexus-toolkit", "nexus-mcp"]
    }
  }
}copy

Manual update

Quick Start

Design-to-Code — generate a full Next.js app from a description

$ nexus run prompt "A SaaS dashboard with sidebar, stats cards, and a data table" \
    -g nextjs-fullstack \
    -p my-dashboardcopy

Nexus will ingest → remap → transform (via Claude) → validate → package — then output a ZIP ready to run.

$ unzip /tmp/nexus-my-dashboard/my-dashboard.zip
$ cd my-dashboard
$ pnpm install && pnpm dev

Dev Agents — run a security review on a file

$ nexus agent run security src/api/routes.tscopy

Workflow Commands — review a PR diff end-to-end

$ git diff main...HEAD | nexus workflow run workflow-review-code -copy

Design-to-Code — from a text prompt

No Figma Make file? No problem. Describe what you want to build:

$ nexus run prompt "A landing page with hero, features grid, pricing, and footer" \
    --golden-path nextjs-static \
    --project-name my-landing \
    --output-dir ~/Desktop/copy

Design-to-Code — from a Figma Make ZIP

Export your project from Figma Make, then run:

$ nexus run zip ~/Downloads/figma-export.zip \
    --golden-path nextjs-fullstack \
    --project-name my-app \
    --output-dir ~/Desktop/copy

Design-to-Code — from an existing codebase

Migrate an existing React/Next.js project to a golden path convention:

$ nexus run codebase ~/Projects/old-react-app \
    --golden-path nextjs-fullstack \
    --project-name migrated-app \
    --output-dir ~/Desktop/copy

Dev-Workflow Agents

22 specialized Claude agents covering every stage of the SDLC. Each agent has deep domain expertise — pass it any content (file, diff, description) and get structured findings back.

List available agents

$ nexus agent listcopy

Run an agent against a file

# Security review on an API route
$ nexus agent run security src/api/auth.ts

# Code review on a PR diff (stdin)
$ git diff HEAD~1 | nexus agent run code-reviewer -

# Deployment review with inline context
$ nexus agent run deployment --context "$(cat docker-compose.yml)"

# Database migration review
$ nexus agent run database prisma/migrations/20240101_add_users.sql

# Architecture review
$ nexus agent run software-architect --context "$(cat ARCHITECTURE.md)"copy

Model selection

By default, agents use claude-sonnet-4-6. Override per-run or set a global default:

# Override model for a single run
$ nexus agent run code-reviewer src/api/auth.ts --model claude-opus-4-6

# Set default model for all runs (add to ~/.zshrc or ~/.bashrc)
$ export NEXUS_DEFAULT_MODEL=claude-opus-4-6copy

Remote server

Run an agent on a remote Nexus MCP server (e.g. a shared team server) instead of the local claude CLI:

# Run code-reviewer on remote server
$ nexus agent run code-reviewer src/api/routes.ts \
    --server http://your-server:3900/mcp

# Inline context
$ nexus agent run product-manager \
    --context "Create a PRD for user auth with OAuth2" \
    --server http://your-server:3900/mcpcopy

When --server is used, the local claude CLI is not required — the agent runs entirely on the remote server.

Agent Memory (v2.3.4)

Agents are stateless by default — every run starts fresh. Enable persistent memory so agents accumulate project-specific knowledge over time, becoming smarter the more you use them.

Enable memory for a run

# Run with memory — inject past findings as context
$ nexus agent run code-reviewer src/api/auth.ts \
    --memory-path .claude/memory/agents/

# Run and save findings back to memory
$ nexus agent run code-reviewer src/api/auth.ts \
    --memory-path .claude/memory/agents/ \
    --remember

# Auto-detect project root (works if you're inside a git repo)
$ nexus agent run security src/api/auth.ts --remembercopy

Memory file layout

Memory lives inside your project — not inside nexus-mcp:

.claude/memory/agents/
├── shared/           # git-committed — team-owned, reviewed in PRs
│   └── code-reviewer.md
└── personal/         # gitignored automatically — per-developer
    └── code-reviewer.md

The personal/ directory is added to your project's .gitignore automatically on first write. Promote curated findings to shared/ via a normal git commit when they're worth sharing with the team.

Memory sections

MCP tools for memory

// Read agent memory
get_agent_memory({ agent_name: "code-reviewer", memory_path: ".claude/memory/agents/" })

// Write / merge findings into memory
update_agent_memory({ agent_name: "code-reviewer", memory_path: "...", content: "...", mode: "append" })

// Modes: append (default) | replace | resetcopy

Workflow Commands

Multi-step orchestration workflows that guide Claude through a structured sequence of tasks. Unlike agents (which are role-based system prompts), workflows are user-level step sequences — design → implement → review → deploy → document.

Available workflows

Run a workflow

# Full code review workflow on a file
$ nexus workflow run workflow-review-code src/api/routes.ts

# Security workflow on a PR diff
$ git diff main...HEAD | nexus workflow run workflow-review-security -

# Deploy workflow with inline context
$ nexus workflow run workflow-deploy --context "$(cat docker-compose.yml)"

# Generate docs for a module
$ nexus workflow run workflow-write-docs src/lib/auth.tscopy

Golden Paths

A golden path is the target stack and conventions Nexus generates code for in the Design-to-Code pipeline.

All Commands

Options