OCR + layout extraction. POST { imageUrl, instruction? }. Returns verbatim transcribed text in reading order, any detected tables as markdown, the primary language, and a handwriting flag. For reading receipts, forms, screenshots, scanned pages, and labels. JPEG/PNG/GIF/WebP.
/api/ai/ocrPAYMENT-SIGNATURE.The ocr API is a pay-per-call ai endpoint built for AI agents and autonomous software. OCR + layout extraction.
There is no signup and no API key. An agent (or any HTTP client) hits the endpoint, receives an x402 "402 Payment Required" challenge, signs a sub-cent USDC payment on Base or Solana, and retries — the data comes back on the paid request. That makes it a drop-in ocr data source for an agent tool-use loop, an MCP host, or a backend that needs ai data on demand without onboarding to yet another vendor portal.
| Name | Type | Description |
|---|---|---|
imageUrlrequired | string | Public image URL (JPEG/PNG/GIF/WebP). max 2048 chars · uri |
instruction | string | Optional steering, e.g. "only the table" or "preserve line numbers". max 500 chars |
# 1. Probe the endpoint with no auth — receive 402 with PaymentRequirements
curl -sS -X POST 'https://2s.io/api/ai/ocr' \
-H 'Content-Type: application/json' \
-d '{"imageUrl":"https://example.com","instruction":"example"}'
# 2. Sign the EIP-3009 transferWithAuthorization for the advertised price +
# payTo from the 402 envelope, then retry with PAYMENT-SIGNATURE:
curl -sS -X POST 'https://2s.io/api/ai/ocr' \
-H 'Content-Type: application/json' \
-H 'PAYMENT-SIGNATURE: <base64-json-payload>' \
-d '{"imageUrl":"https://example.com","instruction":"example"}'
# Or just use the canonical runner — it handles the whole loop:
# EVM_PRIVATE_KEY=0x... node --env-file=.env.local \
# --experimental-strip-types scripts/x402-pay.ts \
# 'https://2s.io/api/ai/ocr'import { TwoS } from '@2sio/sdk'
const client = new TwoS({
privateKey: process.env.EVM_PRIVATE_KEY as `0x${string}`,
})
const result = await client.ai.ocr({
"imageUrl": "https://example.com",
"instruction": "example"
})
console.log('endpoint:', result.endpoint)
console.log('cost:', result.costUsd, 'USDC')
console.log('tx:', result.settlement?.txHash)
console.log('data:', result.data)import os
from twosio import TwoS
client = TwoS(private_key=os.environ["EVM_PRIVATE_KEY"])
result = client.ai.ocr(imageUrl="https://example.com", instruction="example")
print("endpoint:", result.endpoint)
print("cost:", result.cost_usd, "USDC")
print("tx:", (result.settlement or {}).get("tx_hash"))
print("data:", result.data)// 1. Add @2sio/mcp to your MCP host config (Claude Desktop example below).
// EVM_PRIVATE_KEY funds x402 payments per call.
// claude_desktop_config.json
{
"mcpServers": {
"2sio": {
"command": "npx",
"args": ["-y", "@2sio/mcp"],
"env": { "EVM_PRIVATE_KEY": "0x..." }
}
}
}
// 2. Once the server is running, agents call this tool via standard MCP:
{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": {
"name": "ai.ocr",
"arguments": {
"imageUrl": "https://example.com",
"instruction": "example"
}
}
}| Field | Type | Description |
|---|---|---|
ok | boolean | one of: true |
items | array | |
total | integer | Total matching rows upstream; null when unknown. |
source | object |
{
"ok": true,
"items": [
{
"imageUrl": "example",
"fullText": "example",
"tables": [
"example"
],
"language": "example",
"hasHandwriting": false
}
],
"total": 1,
"source": {
"provider": "example",
"url": "example",
"license": "example"
}
}