VibeServe
Enables generating Docker configuration files (Dockerfile, docker-compose.yml) for containerized deployment of UI projects.
Enables generating Node.js deployment configuration files for running the generated UI projects.
Allows using local LLMs via Ollama (e.g., llama3.2) as an LLM provider for generating UI code and related assets without external API keys.
Allows using OpenAI models (e.g., gpt-4-turbo-preview) as an LLM provider for generating UI architecture, code, reviews, tests, and deployment configs.
Enables generating Vercel deployment configuration files for deploying the generated UI projects.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@VibeServe build a login form with email and password fields"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
VibeServe v1.1
The Ultimate Agentic IDE, Orchestrator, and MCP Backend
What is VibeServe?
VibeServe is the first full-stack, fully-integrated AI development platform. It unifies a React-based AI IDE, a Node.js Orchestrator, and a Python MCP Server into a single cohesive system.
It transforms natural language intent into fully-architected, accessible, production-ready UI code โ powered by a 7-step agentic pipeline, a fully aware context engine, and dynamic human-in-the-loop controls.
๐๏ธ architect โ ๐ป code โ ๐ review โ โ
verify โ ๐ iterate โ ๐งช test โ ๐ deployRelated MCP server: OpenGenUI MCP
Unified Repository Structure
This repository is a monorepo housing the three core pillars of the VibeServe ecosystem:
1. ide/ (The Front-End Client)
The VibeServe IDE (formerly Nexus-Alpha) is a Vite/React application acting as the command center. It features an interactive command palette, real-time agent trajectories, code editors, artifact previews, and full MCP tool visibility. It's fully WCAG AAA accessible and designed for extreme performance.
2. orchestrator/ (The Control Plane)
The VibeServe Orchestrator (formerly CodeNexus) is a high-performance Node.js backend. It coordinates agents, handles WebSockets for real-time streaming, runs deep security audits (AST analysis), and manages the lifecycle of the AI agents and their sandbox environments.
3. mcp/ (The Python Backend)
The original VibeServe MCP Server. This is a production-grade FastMCP server containing 28 specialized tools, SQLite-backed memory, multi-LLM routing with auto-fallback (OpenAI, DeepSeek, OpenRouter, Local Ollama), and direct integration endpoints.
Quickstart
One-Command Startup
# Clone and start everything
git clone https://github.com/ncsound919/VibeServe
cd VibeServe/ide
npm startThis starts:
The Vite dev server (port 3000)
The Hono API server (port 3002)
WebSocket connections for real-time agent streaming
Prerequisites
Node.js 20+
Python 3.10+
pnpm (for the orchestrator)
Setup the Entire Ecosystem
Clone the monorepo:
git clone https://github.com/ncsound919/VibeServe cd VibeServeStart the MCP Server:
cd mcp pip install -e ".[dev]" cp .env.example .env # Start in dev mode or hook into Claude Desktop pytest tests/ -vStart the Orchestrator:
cd ../orchestrator pnpm install pnpm run build pnpm run devStart the IDE:
cd ../ide npm install npm run dev
The system will now be fully interconnected.
Architecture
VibeServe IDE (React / Vite)
โ WebSocket / REST
VibeServe Orchestrator (Node.js / Express)
โ stdio MCP Protocol
VibeServe MCP Server (Python / FastMCP)
โโโ 28 Tools ยท 5 Resources ยท 6 Prompts
โโโ V5 Agentic Pipeline
โโโ LLMRouter (Auto-fallback chain)
โโโ MemoryStore (aiosqlite)
โโโ SchemaValidator (WCAG AAA enforcement)
---
## License
MIT License. See [LICENSE](LICENSE) for details.
---
## Windows Bridge (TypeScript + Hono)
The `vibeserve/ts_bridge/` directory contains a **TypeScript-native Hono HTTP bridge** that
replaces the Python `http_bridge.py` on Windows, avoiding ProactorEventLoop issues with
asyncio servers. It spawns `python -m vibeserve` as a child process and communicates with
the Python MCP server via **stdin/stdout JSON-RPC** (the standard MCP stdio transport).
### Quick Start
```powershell
cd vibeserve\ts_bridge
npm install
.\start-bridge.ps1Or with custom settings:
.\start-bridge.ps1 -Port 9000 -Python "python3.12" -Host "0.0.0.0"Architecture
Client (HTTP / WebSocket)
โ
ts_bridge (Hono + TypeScript)
โ stdin/stdout JSON-RPC
python -m vibeserve (FastMCP stdio server)Endpoints
Method | Path | Description |
GET |
| Health check |
POST |
| List available MCP tools |
POST |
| Call an MCP tool by name |
WS |
| WebSocket for streaming tool calls |
WebSocket Protocol
Connect to ws://host:port/ws and send JSON messages:
{ "type": "tool_call", "name": "vs_memory_get", "arguments": { "workspace_id": "demo" } }
{ "type": "ping" }Testing
cd vibeserve\ts_bridge
npm testAvailable Tools
20 toolsgenerate_ui_specC
Generate a production-ready UI specification with multi-agent critique, WCAG AAA validation, and design system enforcement
Inputs: ctx (required), page_type (required), requirements (required), design_system, +2 more. Returns: object
list_design_systemsC
List available design systems
Inputs: ctx (required). Returns: object
memory_statsC
Get statistics on learned/stored UI specifications
Inputs: ctx (required). Returns: object
validate_ui_specC
Validate a UI specification against design system and WCAG standards
Inputs: ctx (required), specification (required). Returns: object
vibe_architectC
Transform natural language intent into a detailed architecture plan with ADR decisions.
Inputs: ctx (required), intent (required), constraints, context, +1 more. Returns: object
vibe_auditC
Full system audit: backend code quality, security, performance.
Inputs: ctx (required), files (required), requirements. Returns: object
vibe_benchmarkC
Run a benchmarking loop with ASCII graphs.
Inputs: ctx (required), iterations. Returns: object
vibe_build_proC
Full professional build: upgrade design -> architect -> code -> verify.
Inputs: ctx (required), intent (required), template, constraints. Returns: object
vibe_codeC
Generate production code from an architecture plan.
Inputs: ctx (required), intent (required), plan (required), constraints, +2 more. Returns: object
vibe_compressC
Compress JSON to TOON format โ 30-60% token reduction.
Inputs: ctx (required), data (required). Returns: object
vibe_deployC
Generate deployment configs for Vercel, Docker, static hosting.
Inputs: ctx (required), project_name (required), files (required), targets. Returns: object
vibe_designC
Generate a landing page using curated DESIGN.md templates.
Inputs: ctx (required), intent (required), template, constraints. Returns: object
vibe_docsC
Fetch documentation for a framework via Context7.
Inputs: ctx (required), query (required), library. Returns: object
vibe_healthC
System health stats.
Inputs: ctx (required). Returns: object
vibe_iterateC
Continuous improvement loop: critique -> repair -> verify -> repeat.
Inputs: ctx (required), specification (required), requirements (required), max_iterations, +1 more. Returns: object
vibe_previewC
Generate a preview HTML page and Playwright test script.
Inputs: ctx (required), html_content (required), filename. Returns: object
vibe_reviewC
Multi-agent code review from three perspectives.
Inputs: ctx (required), files (required), requirements (required). Returns: object
vibe_testC
Generate comprehensive test suites from source code.
Inputs: ctx (required), files (required), requirements, test_framework. Returns: object
vibe_upgrade_designC
Upgrade a design template with senior-dev production patterns.
Inputs: ctx (required), template. Returns: object
vibe_verifyC
Validate code/specs against WCAG, design system, and code quality.
Inputs: ctx (required), specification, files. Returns: object
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelogโฆ
TDQS
Scored across 20 tools
Most tools have distinct purposes, but some overlap exists (e.g., vibe_audit, vibe_review, and vibe_verify all involve code quality checks). Descriptions help clarify, but ambiguity remains for a few.
The majority follow a 'vibe_<verb>' pattern, but several tools like generate_ui_spec, list_design_systems, memory_stats, and validate_ui_spec break this pattern, mixing conventions.
20 tools is on the higher end but justified for a comprehensive development workflow server covering design, architecture, coding, testing, deployment, and auditing. Some tools could potentially be merged.
The toolset covers major stages of development from specification to deployment and auditing. Missing project setup or config management tools, but core workflows are well-represented.
Maintenance
Related MCP Connectors
MCP server that lets your AI assistant build and deploy apps, hosting each one at a live URL.
MCP server for generating rough-draft project plans from natural-language prompts.
MCP server for building and testing AI agents with multi-model experimentation and insights.
MCP server for AI agents to plan, verify, and deploy Cloudflare-native apps.
Related MCP Servers
- AlicenseCqualityCmaintenanceA Model Context Protocol server that enables AI agents to generate, fetch, and manage UI components through natural language interactions.349 npm7ISC
- AlicenseNot gradedqualityDmaintenanceAn MCP server that turns natural-language descriptions into interactive HTML/JS UI components, leveraging OpenGenerativeUI and AG-UI protocol.12MIT
- AlicenseAqualityDmaintenanceAn intelligent MCP server for Google Stitch that generates production-ready UI from text prompts, with auto-orchestration of design systems, WCAG accessibility, responsive breakpoints, and framework conversion (React, Vue, Svelte).1718 npm9MIT
- AlicenseCqualityDmaintenanceA production-grade MCP server providing an autonomous AI agent swarm, persistent semantic memory, browser automation, multi-model reasoning, and 78+ tools for AI-first testing and development.63MIT