MCP 服务器

EVIDIQ Operator

Enables autonomous AI agents to perform browser-based tasks like login, form filling, and data extraction via natural language, with pay-per-call billing and on-chain settlement.

README

<h1 align="center">EVIDIQ Operator</h1>

Computer Use Infrastructure for Autonomous AI Agents.

Reason · Execute · Return — any agent can drive a real browser, autonomously.

Autonomous agents can already discover each other, negotiate, and pay. What they still lack is a way to act on the web — login to a portal, fill a form, download a document, run a multi-step workflow — without a human pressing the buttons.

EVIDIQ Operator is the computer-use layer for the AI agent economy. Delivered as a remote MCP server, billed per call over x402, listed on OKX.AI.

An LLM (GPT-5.6-Terra running via 0G Compute) plans each action from screenshots and drives a real Chromium browser on dedicated Linux desktop infrastructure. The agent never runs browser code itself — it only describes the next action, and we execute it. Every call returns structured JSON.

What it does

Browser automation — natural-language task → structured JSON result.
Web interaction — login, fill forms, navigate, extract data, download documents.
Multi-step workflows — chained sequences on a single persistent browser session.
Verifiable — every paid call can be anchored on 0G Storage (tamper-evident audit trail).
Pay-per-call — no API keys, no accounts, no subscriptions. Just x402 USDT0.

Use it from any agent

# Live pricing discovery
curl -s https://mcp.evidiq.dev/operator/x402

# Connect the remote MCP server (Claude Code, Cursor, OnchainOS, custom)
claude mcp add --transport http evidiq-operator https://mcp.evidiq.dev/operator/mcp

health, capabilities, supported_targets, and estimate_cost are free. The seven browser tools are pay-per-call over x402: unauthenticated requests receive an HTTP 402 challenge; sign it and retry.

MCP Tools

Tool	Cost	Description
`browser_task`	$0.02 USDT0	Natural-language browser task — LLM plans + executes
`login_and_extract`	$0.02 USDT0	Login to a site + extract data
`fill_form`	$0.02 USDT0	Fill + submit a form
`download_document`	$0.02 USDT0	Download a file from a site
`navigate`	$0.02 USDT0	Go to URL, return screenshot
`screenshot`	$0.02 USDT0	Single snapshot of current page
`multi_step_workflow`	$0.02 USDT0	Chained multi-step browser workflow
`health`	Free	Service health + pool telemetry
`capabilities`	Free	List tools + pricing
`supported_targets`	Free	What sites/workflows are supported
`estimate_cost`	Free	Estimate cost for a task

How it settles

EVIDIQ Operator owns reasoning, browser execution, and verification, and settles on open infrastructure:

0G Compute — GPT-5.6-Terra runs inside a Trusted Execution Environment. The model sees screenshots, decides the next action (click, type, scroll, navigate, extract), and returns structured JSON. The LLM never runs browser code.
Dedicated browser infrastructure — Chromium + Playwright on isolated Linux desktop runtimes, spawned on demand, reused while warm, auto-cleaned between calls.
x402 v2 — per-call settlement (EIP-3009 exact / transferWithAuthorization), gasless for the payer, settled on-chain by the server.
X Layer / USDT0 — the OKX A2MCP settlement token (0x779ded…, 6 decimals). $0.02 = 20000 atomic.
0G Storage — paid calls optionally anchored on 0G mainnet (Aristotle, chain 16661).

Proven on-chain


Amount	`0.02 USDT0` on X Layer (`eip155:196`)
Flow	HTTP 402 → EIP-3009 signature → `transferWithAuthorization` (gasless for the payer)
Engine	GPT-5.6-Terra in 0G Compute (screenshot → action → execute → repeat)
Pricing discovery	`https://mcp.evidiq.dev/operator/x402` (per-tool table, 11 entries)

Architecture

flowchart TB
    subgraph client["AI Agent"]
        A["MCP Client<br/>Claude / Cursor / OnchainOS / custom"]
    end

    subgraph server["EVIDIQ Operator MCP"]
        B["x402 v2 Gate<br/>EIP-3009 · USDT0 · X Layer"]
        C["Tool Router<br/>11 tools · 4 free / 7 paid"]
        D["Agent Loop<br/>plan → act → screenshot → repeat"]
    end

    subgraph ai["0G Compute (GPT-5.6-Terra)"]
        E["GPT-5.6-Terra<br/>vision · reasoning · action planning"]
    end

    subgraph runtime["Browser Runtime"]
        F["Chromium + Playwright<br/>dedicated Linux desktop"]
        G["Target Website / App"]
    end

    subgraph chain["On-chain Settlement"]
        H["X Layer (eip155:196)<br/>USDT0 transferWithAuthorization"]
        I["0G Storage (Aristotle 16661)<br/>tamper-evident anchor"]
    end

    A -->|POST /mcp + PAYMENT-SIGNATURE| B
    B -->|verify + settle| H
    B -->|paid call passes| C
    C --> D
    D <-->|screenshot → action JSON| E
    D -->|click · type · scroll| F
    F -->|render · interact| G
    D -.->|optional anchor| I

    style A fill:#6E56CF,stroke:#6E56CF,color:#fff
    style B fill:#2563EB,stroke:#2563EB,color:#fff
    style C fill:#1a130a,stroke:#1a130a,color:#fff
    style D fill:#1a130a,stroke:#1a130a,color:#fff
    style E fill:#00C2A8,stroke:#00C2A8,color:#fff
    style F fill:#C026D3,stroke:#C026D3,color:#fff
    style G fill:#f4efe4,stroke:#e7dcc7,color:#1a130a
    style H fill:#3CCF4E,stroke:#3CCF4E,color:#fff
    style I fill:#00C2A8,stroke:#00C2A8,color:#fff

<details> <summary>ASCII version (for non-GitHub viewers)</summary>

┌─────────────────────┐
│   🤖 AI Agent       │
│   (MCP Client)      │
└──────────┬──────────┘
           │ POST /mcp + PAYMENT-SIGNATURE
           ▼
┌─────────────────────┐    ┌──────────────────────┐
│  ⚡ x402 v2 Gate    │───▶│  ⛓️ X Layer          │
│  EIP-3009 · USDT0   │    │  USDT0 settle        │
└──────────┬──────────┘    └──────────────────────┘
           │ paid call passes
           ▼
┌─────────────────────┐
│  🧰 Tool Router     │
│  11 tools (4/7)     │
└──────────┬──────────┘
           ▼
┌─────────────────────┐    ┌──────────────────────┐
│  🔁 Agent Loop      │◄──▶│  🔒 0G Compute   │
│  plan → act → shot  │    │  GPT-5.6-Terra             │
└──────────┬──────────┘    └──────────────────────┘
           │ click · type · scroll · navigate
           ▼
┌─────────────────────┐    ┌──────────────────────┐
│  🌐 Browser Runtime │───▶│  🌍 Target Website   │
│  Chromium + PW      │    │  (URL target)        │
└─────────────────────┘    └──────────────────────┘
           │
           └─── optional ──▶ 0G Storage (anchor)

</details>

Self-host

# Build & run (Docker)
docker build -t evidiq-operator .
docker run -d -p 3000:3000 --env-file .env evidiq-operator

# Or from source
npm install && npm run build
PORT=3000 node dist/start-server.js

Endpoint: POST /mcp · Discovery: GET /x402 · Health: GET /

Configuration

# Server
PORT=3000
HOSTNAME=0.0.0.0

# Notary/attest signing key (EVM, EIP-191) — same as EVIDIQ main + Notary
NOTARY_PRIVATE_KEY=0x...
OG_PRIVATE_KEY=0x...

# x402 payment (X Layer mainnet, USDT0)
X402_CHAIN=x-layer
X402_ASSET=0x779ded0c9e1022225f8e0630b35a9b54be713736
X402_PAY_TO=0x2a8efe3093278bb4bd3b2d9c7b5ba992ca4fc9b0
X402_PRICE=20000                  # $0.02 USDT0 (6 decimals)
X402_DOMAIN_NAME=USD₮0
X402_DOMAIN_VERSION=1
X402_SETTLE_KEY=0x...             # Gas-funded X Layer wallet
X402_RPC=https://rpc.xlayer.tech

# 0G Storage (mainnet Aristotle, chain 16661)
OG_STORAGE_RPC=https://evmrpc.0g.ai
OG_STORAGE_INDEXER=https://indexer-storage-turbo.0g.ai

# 0G Compute Router ( GPT-5.6-Terra)
OG_COMPUTE_API_KEY=sk-...
OG_COMPUTE_BASE_URL=https://router-api.0g.ai/v1
OG_COMPUTE_MODEL=gpt-5.6-terra

# Browser runtime (internal — provider is an implementation detail, not exposed in user-facing metadata)
BROWSER_API_KEY=...
BROWSER_TEMPLATE_ID=
BROWSER_RESOLUTION_WIDTH=1024
BROWSER_RESOLUTION_HEIGHT=720
BROWSER_TIMEOUT_MS=300000
BROWSER_IDLE_TTL_MS=600000
BROWSER_POOL_MAX=100

Result structure

{
  "task": "Login to example.com and extract the dashboard balance",
  "success": true,
  "steps": 8,
  "summary": "Logged in as user@example.com, extracted balance: $12,450.00",
  "extractedData": { "balance": "$12,450.00", "account": "premium" },
  "storageRoot": "0x…",
  "storageTx": "0x…",
  "stepLog": [
    { "step": 1, "action": "navigate" },
    { "step": 2, "action": "click" },
    { "step": 3, "action": "type" },
    { "step": 4, "action": "click" },
    { "step": 5, "action": "screenshot" },
    { "step": 6, "action": "extract" },
    { "step": 7, "action": "screenshot" },
    { "step": 8, "action": "done" }
  ]
}

Development

npm install      # install deps
npm run build    # tsc → dist/
npm run dev      # tsx watch server.ts