A
licenseA
qualityB
maintenanceEnables AI agents to control a browser via MCP with structured perception, verified actions, and self-healing capabilities.
13
18 npm
1
MIT
Omotai Runtime
Officialby omotai
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Omotai RuntimeGo to my bank, log in with saved credentials, and show my balance"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Omotai Runtime
Status: pre-alpha. Not ready for use. Nothing here is a security guarantee yet.
An MCP server that sits between an AI agent and the browser. The model proposes, the runtime decides.
Hypothesis
A deterministic runtime between the LLM and the browser can drastically reduce harmful actions under prompt injection and keep credentials entirely out of the model's context, without destroying the success rate on normal tasks. This repository exists to test that hypothesis, measured with omotai/eval.
Related MCP server: Chrome MCP Control Center
Planned components
Component | Responsibility |
Tool server (MCP) | Exposes |
Policy engine | Decides |
Secret vault | Secrets bound to origins, filled directly into the page, never shown to the model |
Network guard | Blocks writes to non-allowlisted origins and any secret leaving to an unbound origin |
Human confirmation | Approval through a channel the model cannot reach |
Audit log | Append-only, hash-chained record of what the agent saw, asked and was allowed to do |
Core rule: policies may only allow actions based on facts that require no interpretation (origins, HTTP method, form target, field types, secret bindings). Semantic classification may only make a decision stricter.
Non-goals
A new browser engine. Chromium via Playwright is the substrate.
A protocol for websites or a new action DSL.
Cross-session browsing memory.
Payments, CAPTCHA solving or bot-detection evasion.
Many concurrent agents, multi-browser adapters or a GUI.
Requests in these areas will be closed with a pointer to this list.
Development
uv sync
uv run ruff check .
uv run pytestLicense
Apache-2.0. Security issues: see SECURITY.
This server cannot be deployed
Maintenance
ActivityMaintained
ResponsivenessNo issues
Related MCP Connectors
MCP enforcement layer that intercepts AI agent actions and blocks rule violations before execution.
Zero-secret MCP gateway for AI agents: risk-scored, audited calls with human-in-the-loop approval.
Human-in-the-loop for AI agents over MCP: durable approvals with a hosted review page & audit trail
A paid remote MCP for AI agent browser approval MCP, built to return verdicts, receipts, usage logs,
Related MCP Servers
- FlicenseCqualityAmaintenanceEnables safe automation of Chrome browser through a local MCP server and Chrome extension, allowing LLMs to control browser tabs, pages, and computer-use actions with permission controls.50-
- AlicenseNot gradedqualityBmaintenanceMCP server that gives AI agents secure, consent-based access to a single browser tab, supporting read-only snapshots, element interaction with per-action approval, and frozen multi-step plans, all audited.MIT
- AlicenseAqualityBmaintenanceMCP server enabling AI agents to control a real Linux browser with live view, human takeover, and safety guardrails.131MIT