Customaise: AI Userscript & Agentscript Manager. Powered by MCP & WebMCP

Custom programmable tools for every open tab. For you, or for your AI.

Your AI writes the tool. Customaise runs it. No API keys, no OAuth, no headless browser. Just the Chrome you're already in. 🟢 ONE EXTENSION. TWO PATHS. UNLIMITED SOLUTIONS. Pick the shape that fits the problem. Stack them on any page. ✅ For you. Reshape any page. Inbox that dims promotions. Feed that kills the sidebar. CRM that sorts itself. Every fix becomes a UserScript in your library, running on your sites, following you across devices. The userscript manager that writes itself. ✅ For your AI. Create any tool. Archive the inbox. Place the trade. Export the lead. Book the flight. AgentScripts turn any page into named tools for browser automation, called in milliseconds through WebMCP, under your own session. 🟢 CONNECT YOUR AGENT Customaise exposes 19 tools through MCP, covering script lifecycle, tab control, browser context, visual DOM targeting, and WebMCP calls with HITL consent. Any MCP-compatible coding agent can call them. Uses your IDE's AI subscription, not ours. Or from a terminal. The same server also runs as a customaise command, so an agent that has a shell but no MCP client of its own can drive the browser too. It is the same bridge: the MCP toggle must be on, you must be signed in, and every consent prompt still applies. 🟢 TEST IT LIKE A USER WOULD Deploy the Browser Agent from chat to verify what you just built. It drives the page with real clicks and keystrokes rather than synthetic events, so canvas apps and event-suspicious sites respond, and an action that cannot be carried out reports failure instead of pretending. It runs in your browser, on your screen, and every WebMCP tool it touches passes the same consent gate as everything else. Built on Gemini Computer Use. 🟢 FREE ON-DEVICE AI Chrome 148+ ships with built-in Gemini Nano via the Prompt API. Pick Nano as the model in Chat, or call it from any script via @grant CM_promptAI. No API key. Your prompts never leave your machine. No per-call cost. Works offline once the model is provisioned. 🟢 BRING YOUR OWN MODEL Two ways to run Customaise on something other than our tokens. Both are Power User. ✅ Your key. Save a Google AI Studio or OpenRouter key on your Account Portal and pick any model either one serves. Encrypted with hardware-backed Cloud KMS, never stored in your browser, billed to that provider and uncapped by us. Hold both at once and each model runs on its own key. ✅ Your hardware. Point Customaise at a model server you run yourself: llama.cpp, Ollama, LM Studio, or vLLM, over localhost or a Tailscale name if the machine is in another room. Your models appear in the Chat picker beside the cloud ones. Prompts go straight from the browser to your server and draw no tokens. 🟢 UNDER YOUR CONTROL Every capability a script or agent uses is declared up front and enforced by the extension. ✅ Allow, prompt, or deny per tool. Sensitive calls wait for your explicit OK. ✅ Approve from the tab, or from your phone if you're away from your desk. ✅ @grant gates the GM_* API surface. @connect allowlists network hosts. SRI covers third-party code in @require and @resource. ✅ Chrome DevTools Access is off by default and resets every Chrome restart. Scripts that declare @grant CM_devtools only reach the debugger when you turn the toggle on; Chrome shows its yellow banner while a session is held. ✅ A script written by an AI agent cannot grant itself a tool that runs without asking. It stops at the same prompt anyone else would see. ✅ Every edit creates a restore point. Unlimited undo. 🟢 SHARE WHAT YOU BUILD Publish any script from your Account Portal. Each one gets a dedicated install page you can share. Public listings, private invite links, or links for teams and clients. Subscribers get automatic updates as you improve. 🟢 PRIVACY FIRST In-browser AI runs on paid Gemini APIs, or on a key you bring, or on a model server you run yourself, or on the on-device Gemini Nano model in Chat (Chrome 148+). The last two send nothing to us for inference: the prompt goes to your own machine and the answer comes back from it. Scripts you write can also call on-device AI directly via `@grant CM_promptAI`. When you connect your IDE through MCP, it uses your IDE's AI subscription instead. Your prompts never train a model. Conversations encrypted on your device with AES-256-GCM. Your key, your control. 🟢 READY WHEN YOU ARE One minute to install. One sentence to make your first fix. 🟢 LINKS Website: https://customaise.com Learn UserScripts: https://customaise.com/learn/userscripts Learn AgentScripts: https://customaise.com/learn/agentscripts Run a local LLM: https://customaise.com/learn/local-models Pricing: https://customaise.com/pricing Report bugs: https://github.com/getcustomaise/feedback/issues Changelog: https://github.com/getcustomaise/feedback/blob/main/CHANGELOG.md Chrome Manifest V3.