Turn Any Website into
Live, Verified MCP Tools for AI Agents
Websites are built for humans, not agents. Web2MCP autonomously navigates forms, infers typed schemas, verifies live responses against saved evidence, and generates hosted & stdio MCP servers in minutes.
{
"jsonrpc": "2.0",
"id": 1,
"result": {
"tools": [
{ "name": "search_news", "category": "search", "riskLevel": "read_only" },
{ "name": "get_item_details", "category": "details", "riskLevel": "read_only" }
],
"protocolVersion": "2024-11-05"
}
}How Web2MCP Works
From a raw human portal to an authenticated Model Context Protocol tool
Autonomous Web Exploration
TinyFish browser agents navigate complex SPAs, exercise search filters, detect form inputs, and inspect dynamic responses with anti-bot resilience.
AI Schema Inference & Safety Gate
LLMs analyze observed fields to synthesize typed JSON Schemas. Consequential actions (bookings, purchases) are classified with human confirmation gates.
Live Validation & Evidence Vault
Synthetic probes test real pages before deployment. Every run captures an immutable DOM hash and payload trace to instantly detect website redesign drift.
Production-Ready MCP Capabilities
Everything AI agents need to operate human websites reliably.
Streamable HTTP Transport
Standard JSON-RPC 2.0 streaming endpoints compliant with the 2024-11-05 & 2025-11-25 MCP specifications.
Evidence Vault Auditing
Every synthetic probe captures execution traces and DOM hashes so you know when a target site changes structure.
Two-Phase Safety Gate
Automatic interception for consequential actions (orders, payments) requiring human token authorization.
1-Click IDE Installers
Install generated servers directly into Cursor, Claude Desktop, and VS Code with a single CLI command.
TinyFish Cloud Automation
Cloud browser agents navigate complex SPAs, solve multi-step forms, and maintain authenticated sessions.
Free Tier & Micro-Pricing
3 free generations as a guest, 10 on signup, and affordable $2 Starter and $20 Pro tiers with generous quotas.
Human-in-the-Loop Safety Gate
Agents are powerful, but consequential operations shouldn't run wild. Any generated tool performing bookings, transactions, or state mutations requires explicit confirmation before executing.