Model Context Protocol (MCP)

MegaNova exposes a native Model Context Protocol server so AI agents — Claude Desktop, the Claude Agent SDK, custom MCP clients, and any tool that speaks streamable HTTP MCP — can call MegaNova's inference surface as first-class tools and browse its catalog through resources.

Endpoint

Transport	URL
Streamable HTTP	https://api.meganova.ai/mcp/

The server is stateless and returns JSON responses. There is no separate SSE endpoint — point your MCP client directly at this URL.

Auth model. Tool calls require an API key passed as the api_key argument on each tool invocation (not as an HTTP header). Resources (catalog discovery) are anonymous — no auth required. Keys must begin with sk-.

What the server exposes

Tools (require api_key)

Each tool proxies to the equivalent REST endpoint and preserves billing, rate-limiting, and access-tier checks.

Tool	Purpose	Key arguments
chat_completions	OpenAI-compatible chat	api_key, model, messages[], temperature, max_tokens
text_completions	Legacy completion	api_key, model, prompt, max_tokens, temperature
generate_image	Text-to-image	api_key, model, prompt, plus model-specific options
generate_video	Async text-to-video / image-to-video kickoff	api_key, model, prompt, resolution, duration
get_video_status	Poll a video generation	api_key, generation_id
text_to_speech	TTS audio synthesis	api_key, model, input, voice
list_voices	Enumerate available TTS voices	api_key
create_embeddings	Embedding vectors	api_key, model, input
list_models	All models callable by the caller	api_key
get_model	Fetch a single model's metadata	api_key, model_id

Resources (anonymous)

Resources surface platform state that doesn't need an API key:

URI	Description
meganova://models/all	Every model in the catalog with its media type
meganova://models/image	Image-generation models only
meganova://models/video	Video-generation models only
meganova://gpu/instances	GPU instances grouped by region and type
meganova://gpu/instances/{region}	GPU instances filtered by region

Prompts

Parameterized prompt templates your agent can invoke:

Prompt	Arguments
character_persona	name, role, personality, backstory
image_generation_prompt	subject, style, mood
video_generation_prompt	subject, action, camera, duration

Connecting from Claude Desktop

Add this entry to your claude_desktop_config.json (or the MCP section of your agent config):

{
  "mcpServers": {
    "meganova": {
      "url": "https://api.meganova.ai/mcp/"
    }
  }
}

Once Claude Desktop reconnects, you'll see MegaNova tools and resources in the tools picker. Paste your API key the first time a tool asks for one.

Connecting from the Claude Agent SDK

from anthropic import Anthropic
from anthropic.types.beta import BetaMcpServerConfig

client = Anthropic()

response = client.beta.messages.create(
    model="claude-opus-4-7",
    max_tokens=2048,
    mcp_servers=[
        BetaMcpServerConfig(
            name="meganova",
            url="https://api.meganova.ai/mcp/",
        ),
    ],
    messages=[{
        "role": "user",
        "content": "Use the MegaNova MCP to list all image models, then pick one and describe what it's best for.",
    }],
)

Calling a tool directly (quick check)

Every MCP client does the handshake for you, but if you want to sanity-check the endpoint with curl, send an initialize request:

curl -X POST https://api.meganova.ai/mcp/ \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "initialize",
    "params": {
      "protocolVersion": "2024-11-05",
      "capabilities": {},
      "clientInfo": {"name": "curl-test", "version": "0.1"}
    }
  }'

A successful handshake returns a result with the server's capabilities and a Mcp-Session-Id header you can reuse for subsequent tools/call, resources/read, etc.

Billing and rate limits

MCP tool calls are billed and rate-limited exactly the same as direct REST calls — the API key you pass determines the tier and per-model daily limit. Usage shows up on the same Inference Usage dashboard and via the usage API.

Related

• API Reference - Inference API

• List Models

• Authentication