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
<p align="center"> <img src="https://raw.githubusercontent.com/evidiq/evidiq/main/assets/evidiq.png" width="88" alt="EVIDIQ" /> </p>
<h1 align="center">EVIDIQ Operator</h1>
<p align="center"><strong>Computer Use Infrastructure for Autonomous AI Agents.</strong></p>
<p align="center"> Reason · Execute · Return — any agent can drive a real browser, autonomously. </p>
<p align="center"> <a href="https://mcp.evidiq.dev/operator/mcp"><img src="https://img.shields.io/badge/MCP%20Server-Live-6E56CF?style=flat-square" alt="MCP Server" /></a> <a href="https://0g.ai"><img src="https://img.shields.io/badge/0G-Compute%20%2B%20Storage-00C2A8?style=flat-square" alt="0G Compute + Storage" /></a> <a href="https://www.oklink.com/xlayer"><img src="https://img.shields.io/badge/X%20Layer-Settlement-3CCF4E?style=flat-square" alt="X Layer" /></a> <a href="https://mcp.evidiq.dev/operator/x402"><img src="https://img.shields.io/badge/x402-pay--per--call-2563EB?style=flat-square" alt="x402" /></a> <a href="https://okx.ai"><img src="https://img.shields.io/badge/OKX.AI-A2MCP-121212?style=flat-square&logo=okx&logoColor=white" alt="OKX.AI A2MCP" /></a> <a href="./LICENSE"><img src="https://img.shields.io/badge/License-MIT-3DA639?style=flat-square" alt="License: MIT" /></a> </p>
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 = 20000atomic. - 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
Links
- Live endpoint — https://mcp.evidiq.dev/operator/mcp
- Pricing discovery — https://mcp.evidiq.dev/operator/x402
- EVIDIQ main — https://github.com/evidiq/evidiq
- EVIDIQ Notary — https://github.com/evidiq/evidiq-notary-mcp
- EVIDIQ skill — https://github.com/evidiq/evidiq-skill
- 0G Labs — https://0g.ai
- x402 Protocol — https://x402.org
- OKX.AI — https://okx.ai
License
MIT © 2026 EVIDIQ — see LICENSE. Part of the EVIDIQ trust + execution layer for the AI agent economy.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
Neon Database
MCP server for interacting with Neon Management API and databases
E2B
Using MCP to run code via e2b.
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.