Skip to main content
Kviria ships a stateless Model Context Protocol server alongside the HTTP endpoint. It exposes one tool - kviria - that wraps POST /v1/agent with the same skill contract, the same decisions, and the same wallet limits. No session state.
The tool’s input and output schemas are the authoritative tool definition - identical to the HTTP endpoint, so anything true of the HTTP contract is true of MCP.

​
Authentication

The wallet sign-in rides the MCP transport’s HTTP headers — exactly how x402 payments travel on MCP. Send SIGN-IN-WITH-X (and PAYMENT-SIGNATURE for top-ups) on the HTTP POSTs to /mcp; the same free 50 turns/day per wallet and per-turn pricing apply. Agents built on x402-aware MCP clients do this natively; a bridge client (see the x402 MCP guide) signs on behalf of hosts like Claude Desktop.

​
Connect a client

Any MCP client that speaks streamable HTTP works. For Claude Desktop, add to the config (Settings -> Developer -> Edit Config):
Other clients (Claude Code, Inspector, custom integrations) accept the same URL. There is nothing to install.

​
What the agent sees

One tool: kviria. The agent sends the user’s words or a structured intent plus their anchor, and receives the same five decisions (resolved, confirm, candidates, notFound, clarify) the HTTP endpoint returns. The narrowing loop - echo the candidates back, bind the pick, pass the confirm answer - works exactly as described in the Quickstart.

​
Notes

  • Stateless: all conversation state is caller-held and echoed, same as the HTTP endpoint. The server keeps nothing between calls.
  • Wallet limits apply per turn, same as the HTTP endpoint.
  • Discovery: the server card is published at https://api.kviria.com/.well-known/mcp/server-card.json; the machine spec at https://api.kviria.com/openapi.json.