← Documentation

EVIDIQ Operator

Computer Use Infrastructure for Autonomous AI Agents. Drive a real browser from natural language — login, fill forms, download documents, extract data, run multi-step workflows. GPT-5.6-Terra via 0G Compute plans each action from screenshots.

Quickstart

Connect the EVIDIQ Operator MCP server to any MCP-capable agent:

claude mcp add --transport http evidiq-operator https://mcp.evidiq.dev/operator/mcp

Or probe the live pricing discovery endpoint:

curl -s https://mcp.evidiq.dev/operator/x402

Prefer a skill file? Fetch the agent-readable EVIDIQ Operator skill:

curl -s https://mcp.evidiq.dev/operator/skill.md

MCP server

The remote MCP endpoint is https://mcp.evidiq.dev/operator/mcp (Streamable HTTP). It exposes eleven tools — seven paid, four free:

• browser_task — natural-language browser task (x402-paid, $0.02)
• login_and_extract — login + extract data ($0.02)
• fill_form — fill + submit a form ($0.02)
• download_document — download a file ($0.02)
• navigate — go to URL + screenshot ($0.02)
• screenshot — single snapshot ($0.02)
• multi_step_workflow — chained workflow ($0.02)
• health — service status + pool (free)
• capabilities — tool catalog + pricing (free)
• supported_targets — supported sites/types (free)
• estimate_cost — pre-call cost estimate (free)

browser_task

Submit a natural-language task description. The LLM (GPT-5.6-Terra via 0G Compute) plans each action from screenshots and drives a real Chromium browser.

{
  "task": "Go to example.com, login as [email protected], and download the latest invoice PDF",
  "startUrl": "https://example.com/login",
  "maxSteps": 25
}

Returns structured JSON:

{
  "task": "Go to example.com, login...",
  "success": true,
  "steps": 12,
  "summary": "Logged in, downloaded invoice-2026-07.pdf (245KB)",
  "extractedData": { "file": "invoice-2026-07.pdf", "size": 250880 },
  "storageTx": "0x…",
  "stepLog": [
    { "step": 1, "action": "navigate" },
    { "step": 2, "action": "click" },
    { "step": 3, "action": "type" },
    ...
    { "step": 12, "action": "done" }
  ]
}

How it works

1. Agent calls browser_task with a natural-language goal.
2. Server spawns (or reuses) a dedicated Chromium browser sandbox.
3. GPT-5.6-Terra (via 0G Compute) receives a screenshot + the task.
4. LLM returns a single structured action: click, type, scroll, navigate, extract, done.
5. Server executes the action on the sandbox (via Playwright).
6. Repeat until done or max steps reached.
7. Return: summary + step log + extracted data + optional 0G anchor.

The LLM never runs browser code — it only describes the next action. The server executes it. This is the standard “computer use” agent loop.

Paying from your agent (x402)

The seven browser tools are metered with x402 v2 (scheme exact, EIP-3009, USDT0 on X Layer / eip155:196). An unpaid call returns HTTP 402; the agent signs a gasless authorization and retries with a PAYMENT-SIGNATURE header. The four free tools need no payment.

// 1. Unpaid call -> HTTP 402 with payment requirements
const { accepts: [req] } = await (await callBrowserTask()).json();

// 2. Sign EIP-3009 transferWithAuthorization (gasless for the payer)
const authorization = {
  from: account.address, to: req.payTo, value: req.amount,
  validAfter: "0", validBefore: String(now + 600), nonce: randomHex32(),
};
const signature = await account.signTypedData({
  domain: { name: req.extra.name, version: req.extra.version,
    chainId: Number(req.network.split(":")[1]), verifyingContract: req.asset },
  types: { TransferWithAuthorization: EIP3009_TYPES },
  primaryType: "TransferWithAuthorization", message: authorization,
});

// 3. Retry with PAYMENT-SIGNATURE -> server settles, returns result
const paid = await callBrowserTask({
  "PAYMENT-SIGNATURE": base64({ x402Version: 2, accepted: req,
    payload: { signature, authorization } }),
});

Per-tool pricing lives at https://mcp.evidiq.dev/operator/x402.

Pricing

← Back to docs