Product

One contract across MCP, REST and CLI

A human, a script and an autonomous agent drive Ripar the same way. Connect over MCP and an agent can search, price, pay for and call anything on the network without you writing an integration.

{
  "mcpServers": {
    "ripar": {
      "command": "npx",
      "args": ["-y", "@ripar/mcp"],
      "env": { "RIPAR_MAX_SPEND_PER_DAY": "5.00" }
    }
  }
}

What it does

MCP server

search, call, deploy, workflow.create, jobs.create and receipts, exposed as tools an agent can reach for on its own.

REST API

Cursor-paginated, bearer-authenticated for management. Calling a paid endpoint needs no token at all — it needs a payment.

CLI

init, deploy, logs, rollback, search, call, receipts. Everything the dashboard does, scriptable.

Typed SDKs

TypeScript and Python clients that perform the 402 handshake for you and refuse a quote above your maxPrice.

Spend caps live at the connection

RIPAR_MAX_SPEND_PER_DAY is set where the agent connects and enforced server-side. It is the difference between an agent that buys a service and an agent that drains a wallet because something quoted 40 USDC.

Questions

Which surface should an agent use?
MCP. It is the only one where discovery, pricing and payment are all tools the model can reach for without an integration written ahead of time.
Is the REST API stable?
Endpoints are versioned under /v1 and paginate by cursor. Offsets are not supported because they skip rows when new records land mid-scan.
Do the SDKs handle retries?
Yes, with idempotency keys attached, so a retried call after a network blip does not pay twice.