MCP Servers
The Model Context Protocol (MCP) is how AI assistants connect to external tools. AI-Flow exposes two MCP servers, covering the two halves of working with a workflow.
| Runner | Builder | |
|---|---|---|
| What it does | Runs workflows you already have | Creates and edits workflows |
| Typical ask | "Generate ten product shots with my image flow" | "Build me a flow that summarizes a PDF and posts to Slack" |
| Status | Available | In development |
| Costs credits | Yes, on every run | No — building and saving are free |
| Requires | Tier 3 or above | An account |
They are complementary: the Builder produces a workflow, the Runner executes it. Neither replaces the visual editor — you can always open a flow in AI-Flow and keep working on it by hand.
Why go through a workflow at all?
An assistant can already call models directly. Putting a workflow in between buys you three things it cannot get on its own:
- Repeatability. The flow is a fixed pipeline. The same input gives the same shape of output, every time, whoever runs it.
- Your configuration. Model choices, prompts, chained steps and your stored provider keys live in the flow, not in the conversation. You change the flow once; every caller gets the new behaviour.
- One place to look. Runs, results, expenses and retained media are visible in the app, not scattered across chat histories.
Which one do I want?
Start with the Runner if you already have workflows and want an assistant to operate them.
The Builder is for going from an idea to a working flow without opening the editor. It is not available yet — the page describes what it will do.
A note on credits
Only the Runner spends money, and only when it executes a flow. Listing your flows, inspecting a schema or reading a past result is free.
Assistants are eager. Ours ships instructions telling it to confirm with you before spending, but that is guidance, not a hard limit — an assistant that ignores it will still spend. Keep an eye on unattended loops, and see Costs and safety.