Agent benchmark

Claims about being “agent-ready” are cheap; transcripts aren’t. This page is a real, reproducible run of the core loop against sandbox mode — every payload below is captured verbatim from the API (re-verified 2026-08-28). Result: quote → send → track completes in 3 REST calls (2 MCP tool calls: quote_letter, send_letter). Zero schema retries. Replays provably cannot double-send.

Call 1 — quote (free, mints the confirmation token)

Request

POST /v1/quotes
{ "mail_class": "certified", "page_count": 1,
  "to": { "name": "Property LLC", "line1": "1 Main St",
          "city": "Richmond", "state": "VA", "zip": "23220" },
  "text": "Formal demand for return of my security deposit within 14 days of this notice." }

Response (200)

{
  "status": "ok",
  "total_cents": 1299,
  "total": "$12.99",
  "breakdown": [
    { "id": "certified_mail_letter_1_page", "label": "Certified Mail letter, 1 page", "amount_cents": 1299 }
  ],
  "confirmation_token": "ppq_eyJ2IjoxLCJqdGkiOiJhNDBlZjFiMzFhZDMxYjMx…",
  "expires_in_minutes": 30,
  "next": ["POST /v1/orders with this confirmation_token and the same recipient, content, and options."]
}

Note the next array: the response tells the agent its own next step.

Call 2 — send (sandbox), with an idempotency key

Request

POST /v1/orders
Idempotency-Key: bench-2026-08-25-a
{ "mail_class": "certified", "sandbox": true,
  "confirmation_token": "ppq_eyJ2IjoxLCJqdGkiOiJhNDBlZjFiMzFhZDMxYjMx…",
  "to":   { "name": "Property LLC", "line1": "1 Main St", "city": "Richmond", "state": "VA", "zip": "23220" },
  "from": { "name": "Alex Rivera", "line1": "12 Grove Ave", "city": "Richmond", "state": "VA", "zip": "23221" },
  "text": "Formal demand for return of my security deposit within 14 days of this notice." }

Response (201) — trimmed

{
  "status": "ok",
  "order": {
    "id": "ord_test_f5ca66ace638fdc1",
    "status": "submitted",
    "sandbox": true,
    "price_cents": 1299,
    "tracking_number": "9400MOCK00000001"
  },
  "corrected_to": { "name": "Property LLC", "line1": "1 MAIN ST", "city": "RICHMOND", "state": "VA", "zip": "23220" },
  "capability": { "cancel_token": "…", "review_token": "…" }
}

The address came back USPS-standardized (corrected_to), and the response carries scoped capability tokens for cancel/review — no account, no API key.

Call 3 — track

GET /v1/orders/ord_test_f5ca66ace638fdc1 → 200

{ "status": "ok", "order": { "status": "submitted", "tracking_number": "9400MOCK00000001", "…": "…" } }

The safety proofs (captured in the same run)

Replay the exact same order call (same Idempotency-Key):

{ "status": "ok", "replayed": true, "order": { "id": "ord_test_f5ca66ace638fdc1", "…": "…" } }

Same order id. A retry storm mails one letter. Reuse the confirmation token (new order, same token):

409 Conflict

{ "status": "action_required", "code": "confirmation_used",
  "reason": "This confirmation_token was already used to send a letter. Each quote authorises exactly one letter.",
  "next": ["Quote again — POST /v1/quotes (MCP: quote_letter) — to authorise another letter."] }

Tamper with the token:

400 Bad Request

{ "status": "action_required", "code": "confirmation_malformed",
  "reason": "confirmation_token is not a paperplane quote token.",
  "next": ["Quote first: POST /v1/quotes (MCP: quote_letter) with the exact parameters you intend to send, then pass the confirmation_token it returns."] }

Every failure is a { status, code, reason, next[] } envelope — the error contract documents all of them with retry semantics.

Reproduce it

Run the three calls above against https://sendpaperplane.com with "sandbox": true — no account, no key, no cost. The MCP equivalent is two tool calls (quote_letter → send_letter); connect per Build with agents.

Get started

Guides

Tutorials

Call 1 — quote (free, mints the confirmation token)

Call 2 — send (sandbox), with an idempotency key

Call 3 — track

The safety proofs (captured in the same run)

Reproduce it

​Call 1 — quote (free, mints the confirmation token)

​Call 2 — send (sandbox), with an idempotency key

​Call 3 — track

​The safety proofs (captured in the same run)

​Reproduce it

Call 1 — quote (free, mints the confirmation token)

Call 2 — send (sandbox), with an idempotency key

Call 3 — track

The safety proofs (captured in the same run)

Reproduce it