# Customaise

> Customaise is a Chrome extension for custom tools on the websites you use. UserScripts add controls for you; AgentScripts expose reusable actions to your AI agent through WebMCP, MCP or the CLI. Built by CNDLE Studio LTD. See the pricing page for current plans and allowances.

## Two kinds of custom tools

- **UserScripts:** Add controls, shortcuts, filters or other changes to a page. Customaise provides AI-assisted editing, a script library and portable script files. Generated code still needs review and testing.
- **AgentScripts:** Expose named, site-specific tools with explicit inputs. A compatible AI agent can discover and call those tools through WebMCP, MCP or the Customaise CLI. The tool runs in the browser session where the extension and script are installed.

One workflow can support both: a person can click a page control, while an agent calls a tool for a related action. Web scraping is one example, alongside research, dashboard checks and permitted updates.

## Native WebMCP and connectors

Site-authored WebMCP exposes the actions a website chooses to provide. AgentScripts let a user create additional tools for their own workflows, within their existing access and permissions. These approaches can complement each other. A native connector may already be the simpler solution; a site's current API coverage does not establish what it will offer in future.

## AgentScripts and computer use

Screenshot-driven computer use interprets the rendered interface and acts through browser input. An AgentScript runs saved code for a known action. It can reduce the need to rediscover the same controls on each run, while computer use can help with unfamiliar pages and one-off exploration.

That architectural difference is not a measured speed, price or reliability advantage. Results depend on the model, website, tool implementation, network, retries and approvals. Include script creation, testing and maintenance when evaluating total effort. Both approaches can fail when pages or sessions change; a stable tool name does not guarantee a correct implementation.

No cross-product benchmark is published in this index. Use a representative task and verify its outputs before making a performance claim about Grok Bot, Dots, Muse or another agent.

## Script permissions and review

Review the source and permissions of a third-party script before enabling it. Shared scripts are installed disabled. AgentScript tool permissions distinguish allow, prompt and deny; the selected approval mode and saved decisions determine whether Prompt tools ask again; Full access permits eligible calls. Permissions are not a guarantee that a script is safe or its result correct.

The agent's safeguards, the website's access controls and Customaise's script permissions serve different roles. A working MCP or CLI connection does not override the host's restrictions. See the AgentScripts guide and each Marketplace listing for the relevant controls and scope.

## Knowledge Base

- [What are AgentScripts?](https://customaise.com/learn/agentscripts): User-built WebMCP tools, the `@webmcp` allow/prompt/deny permission model, the HITL consent gate, the two call paths (MCP Server bridge or direct document.modelContext), and how user-authored tools can complement site-provided WebMCP.
- [What are UserScripts?](https://customaise.com/learn/userscripts): The file format, the GM_* API (23 APIs), Customaise's AI editor that writes UserScripts from plain-language prompts, and the portable .user.js export path.
- [Run a local LLM in Chrome](https://customaise.com/learn/local-models): Setup guide for pointing Customaise at a model server the user runs themselves. Covers llama.cpp, Ollama, LM Studio and vLLM; the address policy (loopback, RFC1918, mDNS, CGNAT and Tailscale MagicDNS accepted, public hosts refused); the Ollama CORS flag; the context-window pre-flight; and why prompts on this path never reach Customaise infrastructure for inference and are therefore never metered (with the one caveat that cross-device sync, if enabled, still stores the conversation end-to-end encrypted). Also covers BYOK: a Google AI Studio or OpenRouter key saved on the Account page, encrypted with Cloud KMS, auto-deleted 7 days after last renewal, billed by the provider and uncapped by Customaise.
- [Customaise MCP](https://customaise.com/learn/mcp): The MCP bridge in depth: the two-browser model (agent brain in the IDE, hands in the user's signed-in Chrome), 21 built-in tools, conventions resources agents read before writing scripts, HITL consent on prompt-gated tools, remote approvals from a phone, and the .mcpb bundle for Claude Desktop. The server (@customaise/mcp 3.2.9) speaks the MCP 2026-07-28 revision, negotiated per connection: stateless flow, per-request _meta, cacheable tool lists (1h TTL, cacheScope private); clients on the 2025 revisions work unchanged. Context tools spool large payloads to workspace files for IDE agents; a chat client with no file tools passes output: "inline" (or the install sets CUSTOMAISE_MCP_OUTPUT=inline, as the .mcpb bundle does) and the payload, or for take_screenshot the image itself, arrives in the response.
- [Grok Bot + Customaise](https://customaise.com/learn/grok-bot): Install Customaise in Chrome on the Bot’s cloud computer, then use the CLI from that same computer. Covers sign-in, Agent Bridge, tool discovery, human approvals and a one-time YouTube briefing example before scheduling. The Bot’s browser session is separate from the user’s laptop.
- [The Customaise CLI](https://customaise.com/learn/cli): The customaise command, shipped with @customaise/mcp 3.0.0. Same 21 tools from a shell: doctor (bridge, sign-in, tier, quota, master gate, costs nothing), scripts lifecycle, tab control, WebMCP calls that wait on the browser consent modal, context and screenshots. JSON on stdout so it pipes into jq; exit codes 0-8 are the interface (3 unreachable, 5 cap, 6 consent denied, 7 consent expired, 8 rejected with diagnostics). A resident daemon holds the connection; customaise init writes an AGENTS.md primer so the next agent in the repo discovers the CLI unaided. Same quota and gates as the MCP door: MCP toggle on, signed in, consent in the browser.
- [The Browser Agent](https://customaise.com/learn/browser-agent): Deployed from Customaise chat to verify scripts on the live page. Sends real trusted browser input rather than synthetic events, so canvas apps and event-suspicious sites respond; double-click and right-click behave literally; an impossible action reports failure rather than success. Built on Gemini Computer Use with the selected chat model driving the run, so deploying is refused (with the reason shown) while an OpenRouter model, a local model, or Nano is selected. WebMCP tool calls made during a run pass the same consent gate as every other caller.
- [The marketplace, for agents](https://customaise.com/learn/marketplace-for-agents): How AI agents find and install community scripts through the marketplace’s WebMCP tools, and how you review each script before enabling it.
- [Customaise Knowledge Base hub](https://customaise.com/learn)

## Product

- [How it works](https://customaise.com/features): AI editor, visual DOM selection, cross-device sync, MCP Server bridge, Remote HITL Approvals, Cloud Script Publishing, and on-device AI at script runtime via CM_promptAI.
- [Marketplace](https://customaise.com/marketplace): Curated catalogue of UserScripts and AgentScripts. Community Marketplace listings go through editorial review; review is not a guarantee that code is safe. Per-script trust panel ("Reviewed by Customaise") summarises what each tool reads, what it writes, what external hosts it talks to, and any caveats. Browse by category at /marketplace/category/{productivity, finance, communication, developer-tools, social, other}; per-site landing pages at /marketplace/site/{hostname} aggregate every listing targeting a given site; per-author pages at /marketplace/by/{handle} aggregate every listing from one publisher (handle is the slugified `metadataAuthor` directive: `Jane Doe` becomes `jane-doe`). Per-community landing pages at /marketplace/{cursor, claudecode, mcp} aggregate listings tailored to each agent community. Fork lineage is first-class: every listing carries a "Forked from" chip when applicable, and /marketplace/{slug}/forks lists every fork of a given parent. Subscribe is one click and lands DISABLED in the user's library; the user reviews source, the @webmcp tool surface, and the @connect domains, then explicitly toggles to enable. Pull-update and rollback re-disable the script so the user can review and enable the changed version. Fork is open to every Customaise user (no subscription, no sign-in).
- [Pricing](https://customaise.com/pricing): Free (200K monthly in-browser AI tokens, MCP Server bridge for IDE agents capped at 50 tool calls per UTC day and 150 per rolling 7-day window) and Power User (from $9/month or $72/year, local pricing in GBP/EUR: £7/£60, €9/€72; 5M tokens, unlimited MCP, Cloud Publishing, Remote HITL, BYOK with a Google AI Studio or OpenRouter key, and local models on a server you run yourself).
- [Support](https://customaise.com/support)

## How Customaise compares

These comparisons are published by Customaise. They describe documented capabilities and workflow judgments, not a head-to-head performance or security benchmark. Each article states its review date and links its sources.

### Grok Bot vs Dots vs Muse: browser workflows
URL: https://customaise.com/compare/grok-bot-vs-dots-vs-muse
Reviewed: 30 October 2026

Compare SpaceXSI’s Grok Bot, OpenAI’s Dots and Meta’s Muse for browser tasks, ongoing work and reusable tools. See where Customaise fits, and which connections remain unverified.

Your tools run in your own Chrome, on your own computer, with your sign-ins. Grok Bot uses Customaise through the CLI on its own cloud computer, where Chrome and Customaise run together.

A task can leave behind a tool. You can ask Grok to create an AgentScript, install it with your permission, test its output and repair issues it finds. Review the result before relying on it. Once the tool works, Grok can discover and call the saved action with explicit inputs.

Grok still plans, schedules and interprets the work. Customaise supplies the page-specific execution: collect a table, inspect a dashboard or update a record. It is most useful when a site has no suitable connector, or the existing connector misses the action you need.

This is an extra way to execute browser work, not a claim that the other agents cannot write or reuse code. A native connector may already be the simpler solution.

In Customaise 1.3.9, CM_promptAI runs Gemini Nano on the device, free and with no key, in Chrome 148 or later. Power User adds Script Decisions: focused classification or scoring with Jev or Decisions API, declared in the script with @grant CM_decisions and run with a saved provider key on an authorised device. Provider charges are separate.

Power User can also Send a script through a 24-hour link to an independent editable copy. A signed-in Free recipient can receive it. Copies arrive disabled and remain after the link expires; they do not receive publisher updates. Published subscriptions are a separate sharing option.

We have not verified a Customaise setup for either product. Their documentation does not justify saying integration is impossible. The environment and connection path matter.

Customaise adds setup and maintenance. Its extension must run in the target Chrome profile, with user scripts and Agent Bridge enabled. The CLI needs Node.js 20+ and access to that browser’s local bridge. A login on your laptop does not sign in the Bot’s browser.

Review the script, test its output and keep its permissions appropriate. Website changes can break a tool; expired sessions can interrupt it. Grok’s Bots share an account computer, and installed packages may need attention after recovery.

The agent’s safeguards and Customaise’s tool permissions serve different roles. Adding Customaise does not override the host’s controls or make unrestricted computer access risk-free. Remote approvals from another device require Customaise Power User and the appropriate setup.

Compare the total setup effort and each product’s current allowance. Grok Bot comes with The X Subscription’s Premium, Super and Ultra plans ($30, $100 and $200 a month); Muse is available in the US, Canada and Singapore. Grok Bot access, Dots access and Muse availability depend on their own accounts and rollout terms; Customaise has a separate Free/Power User plan. There is no measured cost or speed winner in this comparison.

Free includes 50 successful Agent Bridge calls per UTC day and 150 per rolling seven days. Power User removes those bridge caps. The external agent’s usage and Script Decisions provider charges are separate. A local bridge does not make optional cloud AI, decision requests, remote approvals or sync local-only.

Sources and scope: https://customaise.com/compare/grok-bot-vs-dots-vs-muse#sources

### Customaise vs AI browsers
URL: https://customaise.com/compare/ai-browser
Reviewed: 2026-10-30

Covers current documented ChatGPT browser tools, Comet and Dia, alongside Customaise 1.3.9. OpenAI retired Atlas on 9 August 2026; its browsing moved into the ChatGPT desktop app and extension. This page does not promise Customaise compatibility inside other browsers.

“AI browser” covers several products and connection models. OpenAI documents both a built-in browser and an extension for existing browsers. Comet and Dia combine their browser experience with assistant features. These are not one uniform architecture or privacy policy.

Customaise is useful when you want a script library for your own page controls and your agent’s named actions. It works in the Chrome you already use, with its sign-ins, and runs your userscripts with full Tampermonkey parity.

**Use Customaise when:** You want to keep browser customisations and reusable agent actions in an extension-managed script library.

**Use AI browsers when:** You prefer a browser’s integrated assistant or ChatGPT’s browser tools and do not need Customaise’s script-management workflow.

**Together:** An assistant and a script manager may complement each other where compatible. Check the actual browser, extension support and connection rather than assuming automatic integration.

Sources: https://customaise.com/compare/ai-browser#sources

### Customaise vs Browser Use
URL: https://customaise.com/compare/browser-use
Reviewed: 2026-10-30

Compares Browser Use’s open-source, Browser Harness and hosted paths with Customaise 1.3.9. Configuration and billing differ between those paths.

The useful distinction is no longer “local browser versus cloud agent.” Browser Use offers several execution paths and agent interfaces, including Browser Harness for coding agents and existing Chrome sessions.

Customaise focuses on the installed script lifecycle: create a page control or named tool, test it, enable it for chosen URLs and keep it available to your agent. Both approaches still need outcome checks and maintenance.

**Use Customaise when:** You want UserScripts for yourself and a managed library of site-specific tools for your external agent in the same extension.

**Use Browser Use when:** You want Browser Use’s automation framework, Harness integration or hosted execution, and are prepared to manage that workflow.

**Together:** You can evaluate both on the same permitted tasks. Measure setup, sign-in, completion quality and recovery as well as model usage.

Sources: https://customaise.com/compare/browser-use#sources

### Customaise vs Browserbase
URL: https://customaise.com/compare/browserbase
Reviewed: 2026-10-30

Browserbase hosting and Stagehand are related but distinct products. This comparison includes Stagehand v4 (4.3, 20 October 2026) and Customaise 1.3.9; local Stagehand does not require Browserbase hosting.

Browserbase supplies hosted browsers and operational tooling. Stagehand provides an automation layer and, in v4, local launches and attachment through a Chromium DevTools endpoint. Its extension supplies the common runtime.

Customaise supplies an installed script library and an MCP/CLI bridge. Choose based on where work needs to run, who maintains it and how people reuse it.

**Use Customaise when:** You want people to install, enable and reuse page-specific controls and agent tools in their connected browser.

**Use Browserbase when:** You need hosted browser infrastructure or prefer Stagehand’s automation APIs and runtime model for your application.

**Together:** A project can use hosted automation for background work and Customaise for browser-side tools. Share workflow requirements rather than assuming identical session access.

Sources: https://customaise.com/compare/browserbase#sources

### Customaise vs Claude in Chrome
URL: https://customaise.com/compare/claude-for-chrome
Reviewed: 2026-10-30

Compares Claude in Chrome’s current documented release and permissions with Customaise 1.3.9. Historical security evaluations are not treated as present-day risk rates.

Claude in Chrome is generally available to paid Claude users according to Anthropic’s August 2026 announcement. It combines browser interaction with Claude’s wider workflow. Customaise is an extension and bridge that a compatible agent can use to write and call installed scripts.

These can be complementary choices. A Claude Code setup can use Customaise’s MCP bridge; that does not mean Claude in Chrome automatically discovers every Customaise tool.

**Use Customaise when:** You want reusable page controls and named browser tools, and you are prepared to test and maintain scripts with your chosen agent.

**Use Claude in Chrome when:** You want integrated Claude browser assistance without first maintaining a separate library of Customaise scripts.

**Together:** Claude Code can use Customaise through MCP. Configure and verify that connection separately from Claude in Chrome.

Sources: https://customaise.com/compare/claude-for-chrome#sources

### Customaise vs Computer Use
URL: https://customaise.com/compare/computer-use
Reviewed: 2026-10-30

Computer Use is a category, not one current model or implementation. This page compares interaction approaches; it makes no category-wide accuracy, latency or cost ranking.

Visual computer use can make unfamiliar interfaces accessible to an agent. Saved browser tools can encode a workflow after you understand the page and its constraints. Some agents combine screenshots, DOM access and tools rather than using a single method.

Customaise supports that loop in the browser: inspect, build, install, test and refine. A saved function remains software that depends on the page and its inputs. Adding an AI decision does not make its output deterministic.

**Use Customaise when:** You have repeated browser work worth encoding and maintaining as a named action or a page control.

**Use Computer Use when:** You need to explore a new interface, operate outside the browser or complete a one-off task where writing a script adds too much setup.

**Together:** Use exploration to understand a permitted workflow, then consider a saved tool for repeated steps. Verify that each tool’s access and outputs fit the task.

Sources: https://customaise.com/compare/computer-use#sources

### Customaise vs Playwright
URL: https://customaise.com/compare/playwright
Reviewed: 2026-10-30

Compares Playwright’s library, CLI and MCP separately with Customaise 1.3.9 / MCP 3.2.8. No matched latency, token or reliability benchmark is claimed.

A Playwright program can accept arguments, branch, expose functions and serve several workflows. Its CLI and MCP server give agents different interfaces to that automation. Customaise goes further inside the browser: AgentScripts add named tools to sites that publish none, each with its own Allow, Prompt or Deny, and the same library gives people page controls.

Customaise packages browser work as installed UserScripts and AgentScripts. The extension manages enablement, URLs, tool permissions and sharing while your chosen agent writes and plans the work.

**Use Customaise when:** You want a library of page controls and named tools that people enable, inspect and reuse from the extension.

**Use Playwright when:** You need browser tests, multiple browser engines, custom automation infrastructure or unattended workflows owned by a software project.

**Together:** A team can use Playwright for automated tests and Customaise for browser-side tools. Validate each workflow in its own execution environment.

Sources: https://customaise.com/compare/playwright#sources

### Customaise vs Claude Skills
URL: https://customaise.com/compare/skills
Reviewed: 2026-10-30

Compares Claude Code skills and host permission controls with Customaise AgentScripts. Other skill hosts have different policies; neither mechanism is a universal sandbox. Anthropic donated the Agent Skills standard to the Linux Foundation’s Agentic AI Foundation on 22 October 2026; more than 70 agent platforms have integrated it.

Skills and AgentScripts are complementary abstractions. Skills organise instructions and may include scripts or other resources. AgentScripts execute in a browser context and expose declared tools. A skill can teach an agent how to use those tools through MCP or CLI.

Claude Code has host permission policies and skill controls. Customaise enforces approval per tool when each one is called. As with a skill, read a script’s code before you enable it.

**Use Customaise when:** You need persistent page controls or named browser operations that people can install, enable and inspect.

**Use Claude Skills when:** You want to teach Claude a procedure, conventions or domain knowledge that can use the host’s existing tools.

**Together:** A skill can describe how to operate Customaise through MCP or CLI. The installed AgentScripts supply the browser actions; both remain subject to their respective controls.

Sources: https://customaise.com/compare/skills#sources

### Customaise vs Tampermonkey
URL: https://customaise.com/compare/tampermonkey
Reviewed: 2026-10-30

Compares Customaise 1.3.9 with Tampermonkey 5.5.1 for Chrome, measured side by side: imports, saved values, settings, GM APIs, network permissions and execution order. Where Customaise differs, it does so on purpose; each difference is listed below.

Both products run userscripts and support agent-assisted editing. Customaise combines page inspection, script writing, installation and testing through MCP or CLI, with AgentScripts for named browser actions.

Tampermonkey offers its own MCP server through the Tampermonkey Editors companion extension and user pairing. It supports listing, reading and editing scripts; newer create and delete operations require Tampermonkey 5.6+ and Editors 1.0.6+. Customaise’s distinction is the integrated browser workflow and agent-callable tool library.

**Use Customaise when:** You want your agent to help create and maintain page controls and reusable browser tools in the same installed library.

**Use Tampermonkey when:** You use Firefox, Safari or Edge. Customaise is built and tested for Chrome.

**Together:** You can evaluate managers alongside each other, but disable duplicate copies so both do not modify the same page.

Sources: https://customaise.com/compare/tampermonkey#sources

## Trust and Legal

- [Privacy Policy](https://customaise.com/privacy)
- [Terms of Service](https://customaise.com/terms)
- [Refund Policy](https://customaise.com/refunds)

## Reference

- [Model Context Protocol (MCP)](https://modelcontextprotocol.io): Anthropic's open standard, published 2024.
- [Chrome Web Store listing](https://chromewebstore.google.com/detail/customaise/anmpijcpaobaabcdncjjmnhdeibipmko)

## Key concepts

- **AgentScript**: A browser script that registers AI-agent-callable tools on a web page via `document.modelContext`, the WebMCP standard name (Customaise 1.4.0 and later; `navigator.modelContext`, its name before Chrome 150, works in every version). Declared with a `// ==AgentScript==` metadata block and `// @webmcp <toolName> <allow|prompt|deny>` directives. Runs in the MAIN world. Designed by Customaise; format is open. Standard file extension: `.agent.js` (parallel to `.user.js` for UserScripts).
- **UserScript**: A JavaScript file run by a UserScript manager on matched URLs. Customaise implements the standard format with an AI editor that writes scripts from plain-English prompts and exports portable `.user.js` files.
- **User-contributed WebMCP**: Tools authored by a user or community as installed AgentScripts. This complements site-authored WebMCP; it does not establish exclusive support for reusable page tools.
- **WebMCP**: The browser-side adaptation of MCP. Shipped in Chrome 146 Canary behind the "WebMCP for testing" flag in February 2026 (joint Google + Microsoft spec, W3C community incubation).
- **@webmcp directive**: Declares the permission for each AgentScript tool. Values: `allow` (autonomous), `prompt` (approval governed by mode and saved decisions), `deny` (blocked). Undeclared tools default to `deny`.
- **HITL consent gate**: The service worker enforces declared tool permissions and selected approval mode. A requested approval shows the tool, arguments and script, and can time out; saved decisions and Full access can permit eligible calls without another prompt. This is not isolation of all script code or whole-computer access.
- **Remote HITL Approvals**: Power User feature that mirrors the consent modal to the user's signed-in account on any device (including phone). Arguments encrypted at rest with Google Cloud KMS.
- **MCP Server (Customaise)**: Local stdio/WebSocket MCP server that exposes registered AgentScript tools to IDE agents (Cursor, Claude Code, OpenAI Codex, Google Antigravity, Amazon Kiro) and MCP-capable desktop agents (Claude Computer Use, ChatGPT Agent). Free for any signed-in user, capped at 50 tool calls per UTC day and 150 per rolling 7-day window. Power User unlocks unlimited MCP. Every successful MCP tool call counts toward the cap; failed calls and protocol traffic do not.
- **@grant CM_devtools**: Opt-in access to the CDP methods and targets Chrome exposes through chrome.debugger. Requires a grant, a justification and the user’s session-scoped DevTools switch. Chrome shows its debugger indicator while attached. Browser restrictions, target/session setup and site behaviour still apply; not every protocol domain, response or input outcome is guaranteed. Logging covers session activity and one-shot calls, not every command within an already-held debugger session. See https://customaise.com/learn/chrome-devtools.
- **Shared script library**: UserScripts provide page controls for people; AgentScripts expose tools for agents. Both are managed in Customaise with script state, URL scope and sharing controls.

## Script sharing (extension 1.3.7)

Quick Actions → Share Script offers Publish & share for subscriptions with publisher updates, or Send for a 24-hour link to an independent editable copy. Sending requires an active Power User; a free signed-in account can receive. A subscribed script must be forked before sending. Each script can have one active link, up to 50 links per account. Copies start disabled, keep their own edits and saved selectors, and remain after expiry or Stop this send, without publisher updates. Reinstalling the same link does not overwrite edits or enable the script. Refresh replaces the active snapshot with the current saved script and resets its 24-hour deadline; opening or copying a link does not extend it. Expired links cannot be revived. Anyone with an active link can preview source. Review source and saved selectors for embedded secrets; saved account keys, GM values, tool approvals and Script Decisions provider choices are not copied. Expiry blocks access immediately, while physical cleanup is asynchronous. Sharing and sync status icons explain each script’s state. See https://customaise.com/learn/agentscripts#publishing. MCP/CLI has no Send or Publish tool.

## Script Decisions (Power User capability)

Available in extension 1.3.8 and MCP/CLI 3.2.7. UserScripts and AgentScripts can ask choice, predicate or score questions about text using Jev or OpenAI Decisions API. Scripts need `@grant CM_decisions` and call `CM_decisions.create` or `createMany` with questions in the Decisions API format, written out in the code. Save separate provider keys on /account#script-decisions, enable the device in /account#devices (from another device when that browser is marked Operated by an AI agent), then select the script's provider from its branch icon or Quick Actions → Script Decisions. Requests run automatically while enabled; Settings → Scripts can pause the device. Data & Analytics in the extension and account shows provider/device usage. Provider charges and budgets are separate from Customaise tokens, and WebMCP tool approval remains independent.

Read https://customaise.com/learn/script-decisions or MCP resource customaise://script-decisions; CLI: `customaise resource script-decisions`. The shipped guide includes a working example and bounded polling/failure handling. Never repeat uncertain paid requests automatically. Quick Setup and /learn/cli provide a copyable agent installation prompt; /learn/grok-bot has the setup prompt for the Bot's own computer.

## UserScript migration and runtime reference (extension 1.4.0, MCP/CLI 3.2.9)

Read https://customaise.com/learn/userscripts#moving-scripts and the handbook
customaise://userscript-conventions (CLI: customaise resource userscript-conventions).
Settings → Scripts → Manual Restore imports Tampermonkey ZIP/JSON backups with
source, saved values and supported settings. Local UserScripts run on Free,
signed out, without AI or Agent Bridge. Preserve imported source and identity;
an IIFE with named functions is a new-code recommendation, not an import rule.

The reference lists 27 GM APIs plus page/window grants. Classic and modern forms
have API-specific returns: legacy GM_cookie methods return undefined and use
callbacks; GM.cookie uses Promises. Related grants can expose API families.
The default Script isolation setting is In the page with a scoped window, not
an isolated world; use unsafeWindow for page variables. Script options can
change defaults. All five @run-at modes are supported; context-menu and
unwrapped execution are supported. Customaise 1.4.0 requires Chrome 147 or later,
matching the tested release baseline.
document-start does not guarantee running before every website script.

Network declarations support IPv6 and port-specific hosts. Under the default
option, specific @connect hosts refuse unlisted hosts; no list or * asks about
each new host. Downloads from unlisted hosts ask first. Responses support
compression and ordered large-body delivery; failure is not truncated success.
Local file dependencies reload on the next page load. Local edits pause automatic
updates. Private windows share temporary values until the last private window
closes. Blocked/Allowed pages and Blocked sources still apply.

Firefox-only features are not provided. Unsupported grants do not prevent
installation; the script runs without those APIs. Imported scripts run
unchanged: the same GM_* APIs and execution order, measured side by side with
Tampermonkey 5.5.1. The public comparison has dated scope and primary sources:
https://customaise.com/compare/tampermonkey.

## Promotional Power User offers

Eligible accounts can explicitly activate three days (72 hours) of Power User
after onboarding. Returning users have a separate one-time offer. Each offer is
once per account; they never overlap, stack, extend an active trial or replace
paid access. Previous welcome/return reward redemptions count for the corresponding
offer. Existing credits and already-active trials keep their original terms.

No card, automatic renewal, bonus tokens or paid 5M monthly allowance is included.
The normal Free token allowance remains. Return to Free at expiry unless the
user chooses a paid subscription; provider and external-agent charges stay
separate. Visiting a link or submitting feedback does not activate an offer.
At expiry, local scripts remain. Downgrade cleanup removes synced cloud copies
and saved provider keys and unpublishes scripts; there is no 90-day cloud grace period.
See https://customaise.com/pricing#trial-details and https://customaise.com/support.

## Marketplace listings (24, served live)

- [Page Scraper](https://customaise.com/marketplace/page-scraper): AgentScript. Turn the lists on any page into rows your AI agent can use: product cards, search results, listings, directory entries, table rows. One call finds the repeating records and the fields in each; another reads every row, following next pages, "load more" buttons and infinite scroll, in your own signed-in session. It can open each row's own page for more detail, save a CSV or JSON file, remember a list as a recipe and report what is new, changed or gone on the next run, and ask your Script Decisions provider which rows match what you want. It only reads: the only things it clicks are Next and "loa Maintained by Customaise.
- [Pain Finder](https://customaise.com/marketplace/pain-finder): AgentScript. Find the problems people describe in a market, on Reddit and Hacker News. Each run reads new posts from the subreddits and searches you choose, plus Hacker News, and asks your Script Decisions provider four questions about each one: does the writer describe an unresolved problem of their own, does the evidence fit your topic and audience, what kind of problem it is, and how strongly they feel it. It returns the problems, strongest first and grouped by kind, with links and the writers' own words, for product research and for writing about what people struggle with. With the same topic and match Runs on: old.reddit.com, www.reddit.com. Maintained by Customaise.
- [Lead Radar](https://customaise.com/marketplace/lead-radar): AgentScript. Find people asking for what you sell, on Reddit and Hacker News. Each run reads new posts from the subreddits and searches you choose, plus Hacker News, and asks your Script Decisions provider three questions about each one: does the writer have a current need your offer meets, is there direct evidence of a fit, and where are they in choosing. It returns the matches with links, so your agent can draft replies for you to post yourself. With the same offer and matching rules, repeat runs reuse earlier judgments. Read-only: it never posts, comments, votes or sends messages. Runs on: old.reddit.com, www.reddit.com. Maintained by Customaise.
- [Reddit Deep Research agent tools](https://customaise.com/marketplace/reddit-deep-research-agent-tools): AgentScript. Reddit research toolkit for AI agents. Six tools: list_posts (search and feed reads with multi-strategy fan-out, a comprehensive deep-research mode with multireddit per-sub fallback and filter auto-relax, plus user-supplied vertical sweep recipes for any domain), get_threads (fetch one or up to ten threads at once with comment triage modes that surface substantive comments raw score missed, sentence-level quote extraction, and cross-thread phrase clustering with optional cross-call entity_bucket tracking), get_user_activity (deep-dive a specific user with optional sub-filter, triage, and quote Runs on: old.reddit.com, www.reddit.com. Maintained by Customaise.
- [Grindset (On-Device AI)](https://customaise.com/marketplace/grindset-on-device-ai): UserScript. Chad is a tiny Stoic Gigachad who lives in the corner of every page. Tell him what you are working on and he coaches you toward it with short, on-device lines from Gemini Nano (CM_promptAI), grounded in real Stoic philosophy: control what you can, love your fate, the obstacle is the way. Stay disciplined and he is the calm, grounded sage. Doomscroll an endless feed like Reddit, X, TikTok, Instagram, YouTube, or Facebook and watch him decay into the anti-Stoic who blames the algorithm, the market, and everyone but himself. A discipline meter tracks it, and you can earn it back with a challenge  Maintained by Customaise.
- [Google Flow agent tools](https://customaise.com/marketplace/google-flow-agent-tools): AgentScript. Drive Google Flow (labs.google) from an AI agent through your own logged-in session. Generate images (returned as a signed URL) and start Veo text-to-video renders from a prompt, poll a render to completion and pull the finished asset, and read your credit balance, your saved agent sessions and their plans, your applets, and likeness eligibility. Runs as you, in your browser, reusing the session the page already holds: no API key to paste, nothing leaves your machine except the same calls the Flow UI itself makes. The two generate tools spend your Flow credits and ask for your confirmation bef Runs on: labs.google. Maintained by Customaise.
- [SEO Audit](https://customaise.com/marketplace/seo-audit): AgentScript. Comprehensive SEO audit for Google ranking. Goes deeper than Lighthouse on the on-page structure, content, and coverage surface: title / meta description (length, truncation, keyword position), canonical URL plus redirect-chain probe (http→https, www, trailing slash), heading hierarchy with order anomalies + duplicate detection, mobile friendliness (viewport, 44px tap targets, 8px spacing, 12px text), E-A-T signals (author, freshness, Organization completeness), featured-snippet readiness (question heading + 10-40-word direct answer), rich-result eligibility per Google's per-type rules (Articl Maintained by Customaise.
- [AEO Audit](https://customaise.com/marketplace/aeo-audit): AgentScript. Comprehensive AEO (Answer Engine Optimization) audit for AI search engines. v2.0 closes the remaining in-browser-checkable gaps against Google's guidance with 12 new advisory checks (all info/minor, gated to stay silent on normal pages): content extractability (long prose with no tables/lists/quotes/stats; question headings with no concise answer beneath; keyword stuffing), E-A-T on article-type pages (missing author/byline; missing or stale published/modified date), page experience (intrusive full-viewport interstitials; heavy ad-slot density), URL quality (opaque non-descriptive slug; multip Maintained by Customaise.
- [iCloud Notes agent tools](https://customaise.com/marketplace/icloud-notes): AgentScript. Typed WebMCP tools for iCloud Notes web. 9 CRUD-shaped tools covering notes, folders, tables, and inline span styling (bold / italic / underline / strikethrough / link). Folders: { list_folders, create_folder, update_folder (rename), delete_folder }. Notes: { list_notes, get_note (returns full body + tables[] inline with rows / cells / uuids), create_note, edit_note (omnibus — title / blocks / body / folderId / tableEdits[] applied sequentially), delete_note }. edit_note covers move + table CRUD (insert / delete / set-cell / set-cells / row + column ops). Body / styling / tables go through App Runs on: www.icloud.com, icloud.com. Maintained by Customaise.
- [iCloud Mail agent tools](https://customaise.com/marketplace/icloud-mail): AgentScript. Typed WebMCP tools for iCloud Mail web. Reads: list_folders, list_threads, list_drafts, get_thread_messages, search_mail. Folder CRUD: create_folder, delete_folder. Drafts: save_draft (basic + reply / forward), delete_draft. Thread state: update_thread (mark read/unread, flag, move, trash, archive, junk). Sending: send_mail, send_draft, schedule_send — all gated by @webmcp `prompt` so the user confirms each send; send_draft + schedule_send additionally require the full envelope + content summary {to, cc, bcc, from, subject, bodyPreview, attachmentCount} and verify each field against iCloud's s Runs on: www.icloud.com, icloud.com. Maintained by Customaise.
- [Google Sheets agent tools](https://customaise.com/marketplace/google-sheets-agent-tools): AgentScript. Drive Google Sheets from an AI agent: read and write cell values and formulas, format cells and borders, clean up data (remove duplicates, split text to columns), build structure (rows, columns, sheets), and add charts, pivot tables, filters, sorting and validation, all inside your own logged-in tab, with no API key and nothing sent to third-party servers. Built on the spreadsheet's own clipboard, toolbar and save surface, with Chrome DevTools used only for the few canvas-only actions the page will not accept as synthetic. Runs on: docs.google.com. Maintained by Customaise.
- [X Deep Research agent tools](https://customaise.com/marketplace/x-deep-research-agent-tools): AgentScript. X (Twitter) research toolkit for AI agents. Read-only tools that run inside your own logged-in x.com tab and turn whatever that tab has open into structured data for your agent: x_page_state (what page you are on and what is ready to read), x_tweet (deep dive on the focused tweet: full text, author, timestamp, engagement, quoted tweet), x_thread (a tweet's reply conversation, scrolled and gathered, with opt-in comment triage, quote extraction, and theme/account clustering — the heart of "what are people saying"), x_search (the open search results for a topic, top or latest), x_profile (a profi Runs on: x.com, twitter.com. Maintained by Customaise.
- [TikTok Deep Research agent tools](https://customaise.com/marketplace/tiktok-deep-research-agent-tools): AgentScript. TikTok research toolkit for AI agents. Eight read-only tools that run inside your own logged-in TikTok tab and turn whatever that tab loads into structured data for your agent: tt_page_state (what page you are on and what is ready to read), tt_video (deep dive on one video read straight from the page data: full caption, hashtags, mentions, engagement, music, duration, and whether it has machine captions, with no extra request), tt_creator (a profile's bio, follower / heart / video counts, verified and region, plus that creator's recent videos), tt_search (videos and creators for a topic from t Runs on: www.tiktok.com. Maintained by Customaise.
- [Instagram Deep Research agent tools](https://customaise.com/marketplace/instagram-deep-research-agent-tools): AgentScript. Instagram research toolkit for AI agents. Eight read-only tools that run inside your own logged-in Instagram tab and return structured data straight to your agent: ig_search (find creators and hashtags for any topic), ig_hashtag (a topic's post volume plus its top and recent posts, with result diagnostics), ig_creator (a profile's bio, follower and post counts, category and links), ig_creator_posts (a creator's recent posts and reels, paginated, filterable to reels only and to a time window), ig_post (deep dive on one post or reel: full caption, hashtags, mentions, engagement, music, location, Runs on: www.instagram.com. Maintained by Customaise.
- [YouTube Deep Research agent tools](https://customaise.com/marketplace/youtube-deep-research-agent-tools): AgentScript. YouTube research toolkit for AI agents, including Shorts. Reads the user's logged-in YouTube session: video meta, timestamped transcripts, comment threads with per-comment like-counts and creator-engagement signals (pinned, hearted, creator-replied), reply trees, search, and channel-video sweeps. Transcripts and comments work for any video or Short by its id, on or off the active tab, so an agent can pull them for search and channel results without opening each one. Engagement-weighted comment ranking maps audience reaction to specific moments in the video via timestamp cross-reference. Contin Runs on: www.youtube.com. Maintained by Customaise.
- [Reddit Thread Triage (On-Device AI)](https://customaise.com/marketplace/reddit-thread-triage-on-device-ai): UserScript. Ambient on-device triage of Reddit comments with Gemini Nano (CM_promptAI). Verdicts are cached for 24h (keyed by comment id + a content hash) so revisiting a thread never re-analyses unchanged comments. When you stop scrolling, it classifies the comments in view (plus about a screen ahead so scrolling stays caught up, top first; a pulsing grey bar marks the one being read), highlights the ones worth reading, tags normal replies, and dims low-effort / bait. Hover any badge for the model's reason. A floating filter (bottom-right) gives live counts and per-category auto-hide toggles (your choice Runs on: www.reddit.com, old.reddit.com. Maintained by Customaise.
- [Reddit Feed Curator (On-Device AI)](https://customaise.com/marketplace/reddit-feed-curator-on-device-ai): UserScript. Ambient on-device curation of the Reddit feed with Gemini Nano (CM_promptAI). Verdicts are cached for 24h (keyed by post id + a content hash) so revisiting a feed never re-analyses unchanged posts. As you scroll, every post in view is tagged plus about a screen ahead so scrolling stays caught up (a pulsing grey bar marks the one being read): the ones worth your attention are highlighted, normal posts are tagged, and low-effort / bait / ads are marked and faded. Hover any badge for the model's reason. A floating filter (bottom-right) gives live counts and per-category auto-hide toggles (your ch Runs on: www.reddit.com. Maintained by Customaise.
- [Quiz Me (On-Device AI)](https://customaise.com/marketplace/quiz-me-on-device-ai): UserScript. Select any text on any page and turn it into instant flashcards and a multiple-choice quiz with on-device Gemini Nano (CM_promptAI). A small button appears by your selection (or press Alt+Q); flip through question and answer cards, or test yourself with multiple choice, build a streak, and get a score. Great for studying notes, articles, documentation, or anything you just read and want to check you understood. Everything runs locally: the selected text never leaves your machine. Free, private, no API key, works offline once the model is provisioned. Maintained by Customaise.
- [Explain This (On-Device AI)](https://customaise.com/marketplace/explain-this-on-device-ai): UserScript. Select any text on any page and get an instant plain-language explanation from on-device Gemini Nano (CM_promptAI). A small Explain button appears by your selection (or press Alt+E) and a panel streams a clear explanation beside it. Switch lens with one click: Explain, Simplify, Key points, ELI5, Define a term, or Translate into your language. Useful for dense legal or medical text, jargon, error messages, or anything you want in clearer words. Everything runs locally: the selected text never leaves your machine, so it works on private pages a cloud tool could never see. Free, private, no API  Maintained by Customaise.
- [Prompt Polish for AI Chats (On-Device AI)](https://customaise.com/marketplace/prompt-polish-for-ai-chats-on-device-ai): UserScript. Polishes your prompt on ChatGPT, Claude, and Gemini with on-device Gemini Nano (CM_promptAI) before you send it. Type as usual; after a short pause a panel beside the input shows a cleaner version, press Tab to accept or Esc to dismiss. Click the sparkle icon next to the send button to manage a library of named instruction sets (e.g. British casual, Code reviews, Marketing voice) that get stacked into every polish, plus tune debounce, style (grammar / clarity / structure), and code preservation per profile. Everything runs locally: your draft never leaves the machine. Free, private, no API key Runs on: chatgpt.com, claude.ai, gemini.google.com. Maintained by Customaise.
- [YouTube Highlights for Your Interests (On-Device AI)](https://customaise.com/marketplace/youtube-highlights-for-your-interests-on-device-ai): UserScript. Reads a YouTube video's transcript and marks the moments worth your time directly on the progress bar, on-device with Gemini Nano (CM_promptAI). Tell it what you care about (pricing, benchmarks, the actual steps, any topic) and it scores each part of the transcript against that, drops a clickable marker on the timeline for every match with the reason on hover, and writes a short summary. Click a marker to jump straight there. Long transcripts are split into sections and read one at a time, with a progress bar while it works; results are cached per video and interest so reopening is instant. Ev Runs on: www.youtube.com. Maintained by Customaise.
- [Binance Sandbox agent tools](https://customaise.com/marketplace/binance-demo-trading-agent): AgentScript. WebMCP trading agent for Binance's demo/paper-trading sandbox. Read-only by default; every write is HITL-gated. Runs on: demo.binance.com. Maintained by Customaise.
- [iCloud Reminders agent tools](https://customaise.com/marketplace/icloud-reminders): AgentScript. Typed WebMCP tools for iCloud Reminders web. 11 CRUD-shaped tools covering reminders, lists, list-groups, time + location alarms, recurrence, tags, early-alerts, URL attachments, smart lists, and subtasks. Lists: { list_lists, create_list, update_list (rename + parent), delete_list }. Reads: { list_reminders (one of listId / view / parentId), get_reminder, get_reminder_attachments }. Writes: { create_reminder, edit_reminder (omnibus — title / notes / move / completed / flagged / priority / due / location / alarm / recurrence / tags / early-alert / urls / clears), delete_reminder, force_delete_ Runs on: www.icloud.com, icloud.com. Maintained by Customaise.
- [iCloud Calendar agent tools](https://customaise.com/marketplace/icloud-calendar): AgentScript. Typed WebMCP tools for iCloud Calendar web. Reads: list_calendars, list_events, get_event. Calendar CRUD: create_calendar, update_calendar, delete_calendar. Event CRUD: create_event (with inline recurrence + attendees + alarms + availability), edit_event (omnibus — title / dates / attendees / recurrence / alarms / availability / location / notes / move), delete_event (scope: this / all / future). Body shapes mirror Apple's wire format (date-array, RFC 5545-ish recurrence, composite occurrence guids); CORS-safe via the text/plain content-type Apple Web uses. KNOWN LIMIT: attendee RSVP responses Runs on: www.icloud.com, icloud.com. Maintained by Customaise.