The comprehensive architectural manual and FAQ for the Supernova Agent Ecosystem.
Unlike simple chatbots, Supernova agents are specialized autonomous workers. They don't just talk; they *do*. Each agent is pre-trained for a specific business vertical and can navigate technical tools to complete complex tasks.
MCP stands for Model Context Protocol. It is an open standard that allows AI models (like Claude, Gemini, or local models) to securely connect to external tools and data sources. Supernova acts as an MCP server, giving your agent access to curated, verified tools instantly.
It's simple: Browse the MCP Tool Gallery (/mcp) or turn your own API into an agent tool in the Creator Portal (/portal). You can invoke tools in Cursor, Claude Desktop, or your agent code via Streamable HTTP, and earn 90% creator revenue via Stripe Connect.
Deploy Now: Perfect for instant tasks. Use it like a pro-tool where you provide instructions and get results in seconds.
Integrate Background: This gives your agent a dedicated endpoint, SDK access, or a Webhook. Use this if you want the agent to automatically handle things as they happen 24/7.
Agents can fail for a few reasons: tool timeout, rate limits, or ambiguous instructions. Check the Execution Logs in your Developer Portal. If an agent loops or times out, refine your prompt to be more specific. If a tool fails, the proxy will return structured error codes explaining the issue.
Free starter tier allows 60 requests per minute (RPM). Paid tiers scale up to 10,000 RPM. Exceeding limits will return a HTTP 429 status code. The MCP connector handles backoff and retries automatically.
Supernova uses deterministic pay-per-call pricing. Callers purchase credit packs via Stripe ($5 for 1,000 credits, $50 for 10,000 credits, where 1 credit = $0.005 USD). Invocations deduct credits based on the creator's defined per-call price. Deductions only commit on verified successful executions; if an upstream tool times out or returns an error, credits are automatically refunded.
Creators earn a 90% revenue share on every tool invocation funded by purchased credits (Supernova retains a 10% platform fee). To receive payouts, connect your bank account via Stripe Connect Express in the Creator Portal. Once your payable revenue reaches $20.00 USD, our daily automated payout engine automatically transfers your earnings to your bank account without manual intervention.
In the Creator Portal, click 'Deploy New Tool'. Enter your REST or OpenAPI endpoint URL, set your price per invocation in USD, and define your input JSON schema. Supernova validates endpoint reachability and SSRF safety, attaches an immutable execution hash log, and registers your tool into the MCP Gateway callable by Cursor, Claude Desktop, and autonomous agents.
Ensure you have Claude Desktop or Cursor configured with an HTTP MCP server. The endpoint is https://supernova.cool/api/v1/mcp/ and requests must send your key in the X-Supernova-Key header. If you are using the local Python connector, ensure you ran pip install supernova-connector and your environment path is correct.
Double-check that you are passing the key correctly. In the MCP config, pass it in the headers object as Authorization: Bearer <key> or X-Supernova-Key. In environment variables, define SUPERNOVA_API_KEY.
If using Cursor or Claude Desktop, restart the application after adding the MCP config. Check the MCP Logs in your IDE. If the connection drops, it usually means your API key is missing or invalid.
Yes. All tool calls are proxied through our secure execution layer with SSRF protection, payload inspection, and heuristic Anomalous Traffic detection. Your data is processed for your task only and is never used to train global AI models. All inputs and outputs are hashed into an immutable audit chain.
Our Execution Proxy automatically scans all incoming payloads and arguments for Anomalous Traffic attacks and exfiltration attempts before the tool ever executes.
The platform tracks live latency and success rates for all registered tools. If a tool becomes degraded, the proxy flags it and halts routing. Furthermore, tools undergo automated health checks via their .well-known/agent.json specs.
