Offline Protocol, Inc.
Offline Protocol is an SDK and platform for devices, apps and machines that coordinate directly with each other: phones, handhelds, gateways and site computers exchange messages, shared documents and service calls over Bluetooth LE, local Wi-Fi and a mesh of nearby devices, with a hosted relay when no local path exists. Teams use it to cut cloud round trips and custom integration work, and to keep field work moving when the uplink is slow or down.
This plugin gives ChatGPT and Codex read-only access to the public Offline Protocol developer registry. Ask which packages fit your app and it looks up SDK packages with their versions, platforms and licenses, product workflows with reference architectures, and integration guides for each area. It resolves the exact install for React Native, React web, Python (PyPI) and Rust (crates.io), explains Bluetooth and local-network permissions, turns an app idea into an architecture plan with a starter template, and previews the files a scaffold would contain.
New in SDK 0.29: the Python service on a Linux or Mac box starts every transport it is configured for, and its HTTP front lets programs that do not use the SDK call a service on another device with an ordinary HTTP request, carried end to end encrypted. Ask the plugin how to run one service per box or how to call a service on another machine, and it answers from the integration guide.
It is for developers building or evaluating apps and systems whose devices must coordinate on site.
Limitations: the plugin is read-only and needs no account. It does not create or edit project files; to create a project or add packages, run the Offline Protocol CLI on your computer (offline create, offline init). It does not read your apps' data, and it does not discover, connect to or control physical devices. Starter templates cover React Native and React web. Native Swift and Android packages are early previews (the Android library is a GitHub release download, not on Maven Central), and the firmware leaf crate compiles for microcontrollers but has not run on a board; the plugin says so rather than presenting them as finished or substituting another framework. There is no Flutter package.
Licenses: the Mesh SDK (React Native, Python, Rust) is AGPL-3.0-only, with a commercial license available from Offline Protocol, Inc. The OfflineID SDKs and Proof of Location (beta) are MIT. Shared types and identicons are ISC.
omg.dev
Describe an app or a game and get a live link you can share. Every app has a database, sign-in and live updates between everyone who opens it, so votes, boards, chats and two-player games work from the first version. Apps are hosted at your-app.omgs.app.
Cafe24 Corp.
Build and deploy web apps through natural conversation on Cafe24 AI SPACE. Create a new site, deploy existing code, and manage projects, environment variables, build logs, backups and access settings. Supports PHP, Node.js, Python and static sites, with MySQL, PostgreSQL and Redis.
한국어 · 대화로 웹앱을 만들고 배포하세요. 기존 코드 배포부터 프로젝트 현황 확인, 환경변수 설정, 빌드 로그 조회, 백업과 접근 제어까지 AI SPACE에서 진행할 수 있습니다.
Español · Crea y publica aplicaciones web mediante una conversación.Français · Créez et déployez des applications web par simple conversation.日本語 · 会話を通じてWebアプリを作成し、デプロイできます。
Requires a Cafe24 AI SPACE account. Files, environment-variable values and database dumps you provide are transmitted to and stored on Cafe24 servers only to build and run your app.Cafe24 AI SPACE 계정이 필요합니다. 제공한 파일, 환경변수 값, 데이터베이스 덤프는 앱을 빌드하고 실행하기 위해 Cafe24 서버로 전송·저장됩니다.
CodexPluginStats
Find Codex plugins by task and source category, inspect their declared skills and archived package files, and explain differences between saved observations with source links. Built for developers and teams evaluating plugins or reviewing changes before use. Requires a CodexPluginStats account and OAuth connection. Read-only access to the saved catalog; no access to your private Following list. Installation counts and revenue are unavailable. Observation dates are not release dates, and archived packages may be incomplete. Independent service, not affiliated with OpenAI.
Akhil Anand
Connect AtlasBurn to see where your AI spend is going across models, providers, workflows, and agents. Ask ChatGPT to break down usage and costs, identify expensive requests or unusual spending patterns, investigate retries and costly agent behavior, and review forecasts and budget guardrails.
AtlasBurn helps teams understand what is driving their AI costs and spot problems before they grow into larger bills.
The connection is read-only and content-blind: it reads cost and usage telemetry from your AtlasBurn workspace, never changes your settings, and never sees your prompts or model outputs.
Expo
Expo helps developers build, test, and deploy iOS and Android apps. Create screens and navigation, connect APIs, add native features, and troubleshoot or upgrade Expo and React Native projects.
Run and inspect your app on an iOS simulator or Android emulator, including cloud devices through Expo Application Services (EAS). Build releases, send beta builds to TestFlight, publish to the App Store and Google Play, deploy web apps, and deliver compatible over-the-air updates.
Includes official Expo guides and tools for projects available to your connected Expo account. Cloud services have plan limits and may incur usage charges. Publishing to app stores requires the relevant Apple or Google developer accounts.
Virtual try-on API & MCP docs
Re8ch
Discover and manage services available to your RE8CH account. Inspect service instances and usage, search reviewed workflows, and plan supported resources before creating them. Sign in with RE8CH; available tools and resources depend on your organization permissions. Platform administration is available only to authorized platform administrators.
Mamdouh Aboammar
Turn a recurring workflow, prompt, or SOP into an agent skill with a clear job, activation boundaries, decision rules, and evaluation cases. OmniSkill routes each request to the next relevant specialist: architecture, authoring, evaluation, repair, host adaptation, or packaging. It checks available host capabilities before using files, shell, or Python, keeps static checks separate from behavioral evidence, and revisits the route when failures or stale verification block progress. Use it in ChatGPT or Codex to build a new skill, audit an existing one, or prepare a portable package. Bundled scripts run only when the host provides the required runtime; no hosted service or MCP connection is required. Cross-host behavior must be tested on each target host before being called verified.
Tekris Labs
Design Game Studio helps create, improve, and test playable 2D, 3D, browser, narrative, and mobile games. It chooses appropriate engines, designs achievable gameplay loops, plans consistent visuals and licensed assets, and guides real implementation, playtesting, optimization, and release handoffs. It works with Phaser, PlayCanvas, Godot, Ren’Py and suitable Unity/Unreal projects. Coding, engine execution, asset creation and multi-agent delegation depend on the tools actually available in the host; this skills-only plugin does not itself supply a game-building cloud service or hosted previews.
Geekflare
Access web data directly from ChatGPT with Geekflare. Search the web, extract public webpage content into structured or LLM-ready results, capture page screenshots, and run website diagnostics such as DNS checks, HTTP header inspection, TLS scans, and performance tests. Useful for research, content extraction, and website troubleshooting. Connect your Geekflare account to authorize requests. Results depend on the availability and accessibility of the target website.
SPENCER DAVID PAULY
Talk to your database with MCP. Securely.
Securely query your PostgreSQL, MySQL, or SQLite database from ChatGPT or Codex. QueryBear sits between your AI agents and your database, giving them read-only access with a security layer you control. Ask questions in plain English, inspect your schema, debug customer issues, and analyze your data.
Connect PostgreSQL, MySQL, or SQLite. Choose which tables and columns your agents can access. Every query passes through table allowlists, column privacy controls, row limits, and timeouts, with a full audit trail. Database credentials are never exposed to the assistant.
Requires a QueryBear account and a connected database or demo database. This hosted plugin cannot write, delete, or modify data. Local and VPN-only databases require the separate local CLI and are outside this hosted plugin's capabilities. Query results are returned to the assistant in your conversation.
WisBot AI agent platform
Parallel Platforms
Investigate website visits with session replay, timed narratives, click maps, funnels and JavaScript error evidence. Trace a snag to its page, device cohort and recorded moment before proposing a fix. Connect your Hoot Lens account and choose the sites and permissions to share. Authorized workflows can annotate findings, preview settings or create a site in your own account. Site creation needs explicit OAuth approval, an available site slot and required visitor consent. Sparse evidence stays qualified; Hoot Lens does not infer visitor opinions, feelings or causation.
Preset
Preset API Skills teaches coding agents to work with Preset and Apache Superset through their public REST APIs: the Preset Management API for teams, workspaces, memberships, invites, roles, and audit logs, and each workspace's Superset API for dashboards, charts, datasets, database connections, SQL Lab, import/export, and embedded analytics.
The package contains 15 focused skills. preset-api covers authentication (exchanging an API token for a short-lived JWT), base URLs, pagination, Rison query parameters, and the shared safety policy; the other skills build on it for workspace discovery, administration, version-aware Superset API use, metadata and data-returning reads, SQL execution, import/export, embedded dashboards, guest tokens, row-level security, and role or permission changes.
Reads run directly with row limits and summarized output. Mutations, imports, exports, credential-bearing reads, guest-token creation, and SQL that is not a confidently classified single-statement SELECT require an explicit confirmation that names the exact target, payload, and expected effect, and credentials and tokens are never printed.
Intended for developers, data engineers, and platform teams who automate Preset or Apache Superset with HTTP requests, curl, or Python.
Limitations: the skills assume a Preset API token is available in the execution environment, and they act only on teams and workspaces that token can already reach. They do not bundle credentials, and they do not create Preset accounts or handle billing.
Sergio Jiménez Rubio
HTML and CSS playground with twenty labs: Grid, Flexbox, Columns, Grid Lanes, border, border-radius, color-mix, box-shadow, datalist, details, Popover, text-box, text-wrap, filter, backdrop-filter, background-image, light-dark(), contrast-color(), transform 2D and dialog. Search by name or purpose, filter by category, adjust native controls, inspect reusable code and continue with the same settings on BaselineLab. Local changes are shared with the model only when requested. Features show current-environment support notices and identify fallbacks; rendering does not establish support in other browsers. Black/white color selection does not certify accessibility.
Build SightLab 2 experiments
Plan a Solana token launch
17811365 Canada Incorporated
Build logging and debugging into your development workflow. LogForge helps you understand why a button fails, where a request breaks, and whether a fix actually works.
Add tagged LogForge.out diagnostics across JavaScript, TypeScript and Python. Follow one feature across the browser, backend and database without enabling every debug message. Read .logforge reports with named checks, failures, unreached outputs and source snippets.
Inspect logs, metrics, inventory and alerts from your connected LogForge instances. In Codex desktop, set up local Docker telemetry, deploy agents to authorized remote Linux hosts, instrument a project and run its checks. In web clients, inspect connected data and receive setup guidance.
Local diagnostic commands work without Docker or an existing appliance. Hosted reads require sign-in and an outbound relay; sharing project reports is explicit. Existing permissions remain in force. Aggregate plugin usage reporting is on by default, with a persistent off switch. Usage reports exclude prompts, source code and log contents.
Charles Chen-Zhou Benjamin
Build and validate MOOS-IvP apps, IvP behaviors, missions, test harnesses, and TIFF background maps. Consult framework documentation, analyze .alog mission logs, and more.
Chan Meng
ArchLang: write floor plans as code, compiled and checked by a deterministic compiler. Skills to author, edit, review and export .arch plans through the archlang CLI, and the ArchLang MCP server. The MCP server is hosted at https://mcp.archlang.uk/mcp with no sign-in: it compiles a plan to an SVG drawing and returns the plan's measured facts and lint findings.
Code changes, drawn simply
GILJOAI LLC
Giljo HQ is a project-management and agent-coordination server for small software teams. It exposes your team's working context (products, projects, tasks, a roadmap, and a durable cross-session memory) over the Model Context Protocol, so your AI coding agent can read real project context and file its own work back into it instead of starting from a blank slate every session. Connect it to ChatGPT or Codex, and your agent can: create and update projects and tasks against your actual roadmap; stage a project for implementation through a human-in-the-loop approval gate before any code is written; record durable memory entries so decisions and context survive across sessions; and coordinate with other agents working the same codebase through a shared message hub. Giljo HQ is a commercial product sold by GILJOAI LLC. Each user signs in with their own Giljo HQ account, and every tool call reads and writes only that user's private workspace.
Work on your paired computers
