meganova-mcp-server 0.3.0

MCP server exposing Nova Mesh agents and skills to Claude Desktop, Cursor, VS Code, and any MCP-compatible client

meganova-mcp-server

MCP server for MegaNova AI — expose LLM inference, image/video generation, embeddings, cloud agents, and Nova Mesh routing to any MCP-compatible client (Claude Desktop, Cursor, VS Code, Windsurf, etc.).

What It Does

MCP Primitive	Capabilities
Tools — Inference	chat_completion, create_embeddings, rerank_documents
Tools — Media	generate_image, generate_video, transcribe_audio
Tools — Models	list_models (with filtering by modality, family, use case)
Tools — Cloud Agents	cloud_agent_info, cloud_agent_chat, cloud_agent_confirm_tool
Tools — Nova Mesh	route_task, chat_with_agent, list_agents, get_agent_info, execute_dag, get_call_log
Tools — Skills	list_skills, execute_skill, search_catalog
Resources	nova://agents (index), nova://agents/{name} (profile)
Prompts	route_task_prompt, agent_chat_prompt, skill_discovery_prompt

Quick Start

Install

pip install meganova-mcp-server

Or from source:

git clone https://github.com/MeganovaAI/meganova-mcp-server.git
cd meganova-mcp-server
pip install -e .

Configure

Set environment variables:

export MEGANOVA_API_KEY=your-api-key                          # Required
export MEGANOVA_API_URL=https://api.meganova.ai/v1            # Default
export MEGANOVA_STUDIO_URL=https://studio-api.meganova.ai     # Default
export NOVA_MESH_URL=https://your-mesh-endpoint               # Optional, for mesh tools

Run (stdio)

meganova-mcp-server

Run (HTTP)

MCP_TRANSPORT=http MCP_PORT=8080 meganova-mcp-server

Client Setup

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "meganova": {
      "command": "meganova-mcp-server",
      "env": {
        "MEGANOVA_API_KEY": "your-api-key"
      }
    }
  }
}

Claude Code

Add to .mcp.json in your project:

{
  "mcpServers": {
    "meganova": {
      "command": "meganova-mcp-server",
      "env": {
        "MEGANOVA_API_KEY": "your-api-key"
      }
    }
  }
}

Cursor / Windsurf

Use the HTTP transport and point to http://localhost:8080/mcp.

Tool Reference

Inference

• chat_completion — Send messages to any MegaNova LLM (Manta, Qwen, Gemini, Deepseek, etc.)
• create_embeddings — Generate text embeddings for semantic search and RAG
• rerank_documents — Rerank documents by relevance to a query

Media

• generate_image — Text-to-image with FLUX, SeedDream, and other models
• generate_video — Text-to-video generation
• transcribe_audio — Audio-to-text transcription from URL

Model Discovery

• list_models — Browse available models with filters (modality, family, pricing)

Cloud Agents

• cloud_agent_info — Get info about a deployed Studio agent
• cloud_agent_chat — Chat with a deployed agent (supports multi-turn conversations)
• cloud_agent_confirm_tool — Approve or reject pending agent tool calls

Nova Mesh

• route_task — Auto-route tasks to the best agent in the mesh
• chat_with_agent — Direct message to a specific mesh agent
• list_agents / get_agent_info — Discover mesh agents
• execute_dag — Run multi-step DAG execution plans
• get_call_log — View recent LLM call history

Skills

• list_skills / search_catalog — Discover available skill packs
• execute_skill — Run a specific tool from a skill pack

Docker

docker build -t meganova-mcp-server .
docker run -e MEGANOVA_API_KEY=key -p 8080:8080 meganova-mcp-server

Development

pip install -e ".[dev]"
pytest

License

MIT