← 返回
未分类 Key

Screenshots for AI agents via x402

Browser screenshots and vision over MCP Streamable HTTP with x402 USDC micropayments at this deployment. Paid tools require an x402-aware MCP client and a fu...
通过MCP流式HTTP提供浏览器截图和视觉功能,支持x402 USDC微支付。付费工具需要x402感知的MCP客户端...
buildinhk
未分类 clawhub v1.0.1 1 版本 99635 Key: 需要
★ 0
Stars
📥 273
下载
💾 0
安装
1
版本
#latest

概述

screenshot-x402 — MCP screenshots + x402

Remote Model Context Protocol server (Streamable HTTP only). All tools are invoked through MCP — there is no separate REST “screenshot URL” for paid capture.

Base URL: https://screenshotx402.com

MCP endpoint: https://screenshotx402.com/mcp

Before you install or use

  1. Read https://screenshotx402.com/discovery.json for list prices, x402 network, and mcp.url — avoid surprise charges.
  2. Confirm your agent host can run an x402-aware MCP client and sign USDC payment authorizations on that network.
  3. Trust this deployment (https://screenshotx402.com) before placing wallet / signing secrets in environment or vaults.
  4. Call health (free) before paid tools.

Credentials and signing (paid tools only)

Free: health and discovery HTTP need no wallet.

Paid (take_screenshot, analyze_screenshot): the client must sign x402 USDC authorizations and retry with payment proof headers (e.g. PAYMENT-SIGNATURE / X-PAYMENT). Use a stack such as Cloudflare agents/x402 withX402Client with a viem account. Private keys belong in your host env (examples in project docs: AGENT_PRIVATE_KEY, X402_PRIVATE_KEY in sample clients) — never in prompts. The signer needs USDC on the advertised network.

If your platform cannot attach payment proofs, use only free discovery + health, or another capture method.

Privacy and off-host data

URLs you submit are sent to this deployment (https://screenshotx402.com). The service fetches and renders pages on operator infrastructure; screenshots and, for analyze_screenshot, your prompt and vision output, are processed off your machine.

Do not use for non-public or sensitive URLs (internal hosts, auth-gated content you cannot leak, secrets in query strings). Use local or self-controlled tooling for private systems.

Discovery (free HTTP)

ResourceURLPurpose
------------------------------------------------------------------------------------------------------------
This skillhttps://screenshotx402.com/skill.mdHuman + agent onboarding (this file)
Machine discoveryhttps://screenshotx402.com/discovery.jsonmcp.url, x402Network, tool list, list prices
API referencehttps://screenshotx402.com/docsFull parameter tables and return shapes
Landinghttps://screenshotx402.com/Tools overview and examples

Always read https://screenshotx402.com/discovery.json for current USD list prices — do not hard-code amounts.

Tools overview

ToolCostPurpose
-------------------------------------------------------------------------------
healthFreeSmoke test MCP + advertised x402 network
take_screenshotx402 / USDCBrowser PNG or JPEG of a public https URL
analyze_screenshotx402 / USDCJPEG capture + vision text answer to your prompt

x402 payment flow (MCP)

This service uses MCP tool calls instead of raw GET /api/..., but the idea matches HTTP x402:

  1. Connect a Streamable HTTP MCP session to https://screenshotx402.com/mcp.
  2. Call a paid tool (take_screenshot / analyze_screenshot) with normal arguments.
  3. First response encodes payment required: tool result includes x402 metadata (version, accepts with price, network, pay-to, asset, facilitator expectations). No image payload yet.
  4. Complete the USDC payment on the chain advertised in that payload (this deployment’s network in discovery is base — confirm in discovery.json / health).
  5. Retry the same callTool with the payment proof attached the way your MCP client expects (e.g. PAYMENT-SIGNATURE / X-PAYMENT on the MCP HTTP session, per your stack).
  6. Server verifies via the facilitator, then returns the real tool result (image and optional text).

Automatic handling: Use an x402-aware MCP client (for example Cloudflare agents/x402 withX402Client around the MCP Client) so payment discovery, signing, and retries are handled like an x402 HTTP client would handle 402 + X-Payment — without you manually copying headers.

1. health (free)

MCP: callTool with name health, arguments: {}.

Parameters: none (empty object).

Successful result (shape):

{
  "content": [
    {
      "type": "text",
      "text": "{ \"ok\": true, \"name\": \"screenshot-x402\", \"x402Network\": \"base\" }"
    }
  ]
}

content[0].text is a JSON string — parse it for ok, name, x402Network.

2. take_screenshot (paid — x402)

MCP: callTooltake_screenshot.

Parameters:

ParamTypeRequiredDefaultDescription
------------------------------------------------------------------------------------------------------------------
urlstring (URL)yesAbsolute https:// page to capture
widthnumberno1920Viewport width (100–3840)
heightnumberno1080Viewport height (100–2160)
fullPagebooleannofalseCapture full scrollable page
delaynumberno0Extra wait after load (ms, max 30000)
cacheTtlnumberno86400R2 cache TTL seconds; 0 skips cache reads
formatstringnopngpng or jpeg
colorSchemestringnono-preferencelightdarkno-preference
deviceScaleFactornumberno1Pixel ratio 1–3 (sharpness)
hideSelectorsstring[]no[]Up to 40 CSS selectors to hide

Successful result (shape):

{
  "content": [
    {
      "type": "image",
      "data": "<base64 PNG or JPEG>",
      "mimeType": "image/png | image/jpeg"
    }
  ],
  "_meta": {
    "cached": true,
    "renderTimeMs": 1234
  }
}

_meta.cached and _meta.renderTimeMs may be omitted depending on path.

3. analyze_screenshot (paid — x402)

MCP: callToolanalyze_screenshot.

Parameters:

ParamTypeRequiredDefaultDescription
---------------------------------------------------------------------------------------------
urlstring (URL)yesPage to capture
promptstringyesQuestion/instruction for the vision model
widthnumberno1920Viewport width
heightnumberno1080Viewport height
fullPagebooleannofalseFull page capture
colorSchemestringnono-preferenceSame as take_screenshot
deviceScaleFactornumberno1Same as take_screenshot
hideSelectorsstring[]no[]Same as take_screenshot

Successful result (shape):

{
  "content": [
    {
      "type": "image",
      "data": "<base64 JPEG>",
      "mimeType": "image/jpeg"
    },
    {
      "type": "text",
      "text": "<vision model answer>"
    }
  ],
  "_meta": {
    "renderTimeMs": 1234
  }
}

Decision guide

GoalToolCost
--------------------------------------------------------------------
Verify MCP + networkhealthFree
Still image capturetake_screenshotPaid (see discovery)
Capture + describe / extractanalyze_screenshotPaid (see discovery)

Constraints

  • Targets must be https:// URLs (see /docs).
  • Vision output for analyze_screenshot depends on the operator’s server configuration (not something callers configure through MCP).

More links

版本历史

共 1 个版本

  • v1.0.1 当前
    2026-05-08 00:09 安全 安全

安全检测

腾讯云安全 (Keen)

安全,无风险
查看报告

腾讯云安全 (Sanbu)

安全,无风险
查看报告

🔗 相关推荐

ai-agent

self-improving agent

pskoett
捕获经验教训、错误及修正内容,以实现持续改进。适用于以下场景:(1)命令或操作意外失败;(2)用户纠正Claude(如“不,那不对……”“实际上……”);(3)用户请求的功能不存在;(4)外部API或工具出现故障;(5)Claude发现自身
★ 4,078 📥 808,672
dev-programming

Github

steipete
使用 `gh` CLI 与 GitHub 交互,通过 `gh issue`、`gh pr`、`gh run` 和 `gh api` 管理议题、PR、CI 运行及高级查询。
★ 676 📥 325,263
ai-agent

Self-Improving + Proactive Agent

ivangdavila
自我反思+自我批评+自我学习+自组织记忆。智能体评估自身工作、发现错误并持续改进。
★ 1,376 📥 320,195