How Toolscaled works
Toolscaled turns a task that you did once with your agent into a workflow that runs again without you. This page puts the whole idea in one place, so that each other page fits into it.
1. Do the task once with your agent
Section titled “1. Do the task once with your agent”You work in the chat that you already use: Claude, ChatGPT, Gemini, Grok or another agent that supports MCP. Through Toolscaled, the agent reaches the accounts and tools that you connected, for example Gmail, a sheet or your CRM. You ask, you read the result, and you correct it until it is right. While you work, Toolscaled records each tool call and its result.
2. Save it as a workflow
Section titled “2. Save it as a workflow”When the result is right, tell your agent: “Turn what we just did into a workflow.” The agent writes the steps in words and asks you what changes from one run to the next, such as a date or an address. Those become the inputs of the workflow.
A step that needs a model to decide or to write, such as a summary, keeps an AI model. Every other step calls its tool directly. This is why a workflow costs less than asking the agent again each time, and why its steps stay the same from run to run.
3. Check it, then turn it on
Section titled “3. Check it, then turn it on”Read the steps before anything runs on its own, and tell the agent what to change. When they are right, approve them, and the agent publishes the workflow.
Then test it. A test run reads your accounts but sends and writes nothing: it uses what your first run in the chat sent and wrote. The agent asks you before a test sends a mail or writes to an account for real. Nothing runs by itself until you turn on its automation.
4. It runs when the work arrives
Section titled “4. It runs when the work arrives”An automation starts the workflow on a schedule, or when a new email, a webhook call or a form entry comes in, or after another workflow. You can also press Run now on the website, or ask your agent to run it. Each start makes a run, and the page of the run shows what each step did and what it cost. A step can also stop and wait for your approval, for example before it sends a mail.
5. When a run fails, fix it in chat
Section titled “5. When a run fails, fix it in chat”A failed run shows in Action items, with the step where it stopped. When the cause is outside the workflow, for example an account that you must connect again, you retry the run from where it stopped. When the workflow itself must change, you open the run in your chat. Your agent sees what happened, fixes the workflow with you, and publishes the new version, so the next run uses the fix.
Where to go next
Section titled “Where to go next”Sign up if you have no account yet. Your first workflow then goes through these five steps with a real example: a short summary of the AI news, each morning.