Updated
Each guide answers one question in its first lines. They are grouped: questions, example workflows you can add and run, and comparisons.
Plain answers to what people ask: what an MCP workflow is, how to make an assistant follow a checklist and the words we use.
An MCP workflow is a process written as steps that an AI assistant works through by calling an MCP server. The server keeps the position and hands over only the current step, so during a run the assistant sees no later step. Pathista is such a server: you write the workflow once and run it in ChatGPT or Claude.
To make ChatGPT follow a checklist, give it one step at a time instead of the whole list, and make it show evidence for each item. A numbered prompt relies on the model's discipline. A workflow server such as Pathista keeps the position outside the chat, shows only the current step and closes items only with evidence.
This glossary defines the words used to run workflows with an AI assistant: workflow, step, edge, condition, run, sub-run, todo, checklist, evidence, note, MCP and MCP connector. Each definition is the one Pathista uses in the app and in the help center, and links to the longer explanation on the concepts page.
Ready-made templates, explained step by step. Each one you can add to your account and run.
A release checklist for Claude Code works best when publishing sits behind recorded evidence. The Pathista template runs the tests on the release commit, loops through fixes for up to three rounds, checks that the tag points at the tested commit and opens the publish step only after your explicit yes.
To turn several customer interviews into insights, make every claim traceable. The Pathista template records one card per interview with word-for-word quotes and their place in the notes, asks you to check a sample against the sources, and counts a theme only when at least two different interviews back it.
An independent critic loop separates building from judging: the assistant builds, a reviewer with a fresh context checks the result against a written bar and names one largest gap, and the work repeats until the bar is met or the round budget is spent. The Pathista Gauntlet loop template runs it with a default budget of three rounds.
Pathista beside other ways of running a process, with every statement about another product sourced.
Pick a Custom GPT when you can write the task down as one set of instructions and you are there to watch it. Use Pathista when a process has steps that must happen in order, loops or a record. A Pathista run keeps its position on a server, shows one step at a time and works in ChatGPT and Claude.
Use a Claude project when you want standing instructions and knowledge, and a Skill when you want know-how to apply when it is relevant. Use Pathista when a process must go step by step with its position kept outside the chat. A Pathista run controls where in a process the assistant is.
Use Zapier or n8n when you want an AI agent that connects your apps. Use Pathista when a person works with an assistant in a chat and the process must go step by step with evidence and a record. A Pathista run guides the assistant you are talking to.
A pasted checklist is enough for a short list you can watch, once. Pathista is better when the list is long, repeated, branches, spans several chats or must leave a record. A pasted prompt gives the assistant the whole list. A Pathista run shows one step and keeps the evidence.