You keep prompting, we convert repeatable workflows into agents.

Autoplay is an MCP connector for Claude and ChatGPT that turns the work you repeat into AI agents. While you prompt as usual, it spots repeatable workflows, saves them as agents, and keeps improving them every time they run, so you get consistent results without rewriting prompts.

An agent is your best prompt for a task you do often, saved with its inputs, steps, output format and rules, and kept up to date for you. Instead of typing the same instructions every time, you just ask for the task, for example “follow up with Acme”, and the agent runs the full workflow.

Skills are instructions you write and maintain yourself, and they only help when you remember to use them. Autoplay is the proactive layer on top: it decides for you when something is worth turning into an agent, which agent fits what you are asking right now, and when an agent needs updating. That part is personal to how you work and cannot be hard-coded. Autoplay agents also keep a full version history and work in both Claude and ChatGPT.

Autoplay acts without being asked. It notices when a task you just did is worth saving, recognises when one of your agents fits a request even if you never mention it, and suggests improvements when your needs change. These moments are different for every person, so Autoplay learns them from your own usage instead of following fixed rules, and it always asks before saving or changing anything.

You can write a Skill or a saved prompt yourself, and for one or two tasks that works well. The hard part to build is the proactive layer: knowing, for you specifically, which requests are worth automating, when an agent should run and when it needs updating. That is not a rule you write once, because it depends on how each person works and changes over time. It is the part Autoplay does for you, across Claude, ChatGPT and your team.

Custom GPTs and prompt libraries require you to build, organise and update everything by hand, and they often end up unused. Autoplay lives inside the chat you already use: it suggests agents when it notices a repeat, runs them automatically when they fit what you ask, and updates them as your work evolves.

Projects are folders for one topic: you add files and custom instructions, and they only apply to chats inside that project. Autoplay agents are reusable workflows that work in any chat, start on their own when a task fits what you are asking, and improve every time they run. The two work well together: Projects keep the background context for a topic, while Autoplay handles the repeatable tasks you do across all of your chats and projects.

No. Autoplay writes and structures the prompt for you based on how you already work. The quality of the result no longer depends on how well you write a prompt each time, because the agent improves with every run.

No. Agents activate on their own when your AI assistant recognises that one fits what you are asking, so you can keep talking naturally. If you prefer shortcuts, you can also call any agent by name, for example @call-follow-up.

Autoplay works with Claude and ChatGPT through the Model Context Protocol (MCP), an open standard for connecting AI assistants to external tools. Your agents can use the tools you have connected to your assistant, such as Gmail, Google Calendar, Slack, Google Drive, Notion or HubSpot.

MCP is an open protocol that lets AI assistants like Claude and ChatGPT connect to external tools and data in a standard way. Autoplay is an MCP connector, so you add it to your assistant once and it works inside your normal chats, with no new app to learn.

Autoplay learns from every run: when your requests start to differ from the agent, or a connected tool changes, it proposes an update. You approve changes with one tap, nothing changes without you, and every earlier version stays available to restore.

Yes. You can share any agent with teammates so everyone runs the same best version, and teammates can propose edits that you review and approve. Agents are private to you until you choose to share them.

Yes. Agents are kept lean, with no repeated or unnecessary context, so each run uses fewer tokens than rewriting a long prompt from scratch while producing the same result. This happens automatically.

Autoplay is SOC 2 Type II and ISO 42001 compliant. Your agents are private by default, changes always require your approval, and you can restore any previous version at any time. Full details are in our Trust Center.

Autoplay is free for up to 10 users, with unlimited agents included. Teams of 10 or more pay $10 per user per month, and Enterprise plans are tailored to your team. You can try it now or book a free demo.

Setup takes a few minutes: add the Autoplay connector to Claude or ChatGPT and keep working as usual. We also offer a free 15-minute onboarding call, and most people have their first agent running before the call ends.