Small model?
Ask a stronger one.

Start here / One tool

Ask. Get help. Continue.

ask_expert({"question":"Check my solution. Here are the task, code, and error: …"})

codex mcp add nuri-expert --url https://router.paymentrequired.com/mcp/expert --bearer-token-env-var ASK_EXPERT_TOKEN

The router selects a configured expert with recent healthy observations. It can try one fallback before issuing an invoice. Unknown payment outcomes never trigger a new charge. No expert available? You receive a clear error, not a payment request.

Browse configured models, inspect rates, and test an actual text response. A configured model may be unavailable or out of credits; API discovery only lists models with recent healthy observations.

Rates are USDC per million tokens, plus the model’s request fee. The published authorization ceiling is 1 USDC per request, not the price. Review the exact invoice before signing.

02 / Advanced access

Choose a model yourself.

1. Connect Codex or your MCP client

   Add the remote server for free model discovery and quotes.

   Codex CLI

   codex mcp add nuri-router --url https://router.paymentrequired.com/mcp

   Other MCP clients: HTTP connection config

   This SDK-neutral connection example uses a common MCP config shape. Adapt the wrapper to your client.

   {
     "mcpServers": {
       "nuri-router": {
         "type": "http",
         "url": "https://router.paymentrequired.com/mcp"
       }
     }
   }

   Connecting does not enable payment. Paid chat requires an x402-aware signer for direct Base USDC, including preservation of the exact invoice resource. A generic MCP client cannot pay on its own.

2. Choose a model and inspect its price

   Call list_models, then quote with an exact model ID. Both are free. A quote is a non-payable estimate; it does not generate an answer.

   Or discover over HTTP — free, no wallet

   curl -sS https://router.paymentrequired.com/v1/models

3. Prepare an answer, then review the invoice

   An unsigned chat request consumes provider compute to prepare and meter an answer. It does not charge the buyer or release the answer. Expect HTTP 402 with an exact invoice in PAYMENT-REQUIRED.

   Unsigned HTTP example — run only when ready

   Select a model above, or replace the placeholder with an ID from the current catalog. This Bash/zsh example needs Node.js to create a private UUID. Keep the same key and body for the paid retry and recovery.

   # Run once per NEW request. Keep this shell open.
   # Keep the key and returned invoice private.
   router_request_id="$(node -p 'crypto.randomUUID()')"
   curl -sS -i https://router.paymentrequired.com/v1/chat/completions \
     -H 'Content-Type: application/json' \
     -H "Idempotency-Key: $router_request_id" \
     --data-raw '{"model":"MODEL_ID_FROM_CATALOG","messages":[{"role":"user","content":"Say hello."}],"max_tokens":64,"stream":false}'

   This command has no payment signature. Do not use unsigned chat as a health check. Running the whole snippet again creates a new request key.

   Have your wallet client verify the invoice’s amount, Base network, USDC asset, recipient and expiry. Approve the exact amount, then retry the identical request with the signed payment and the private invoice resource intact.

03 / Payment & recovery

An invoice for actual usage.

1. 1Prepare

   The provider generates an answer.

2. 2Invoice

   Actual tokens determine the amount.

3. 3Approve

   Your wallet signs the exact invoice.

4. 4Paid retry

   Settlement releases the answer and receipt.

04 / Reference

Small surface. Explicit payment.