Connect the AI You Already Pay For
Polyz speaks MCP, so Claude Code and other MCP clients can work directly inside your manuscript — reading your chapters, proposing edits you review, and writing entire courses into your library. It runs on your AI subscription, not ours.
Three minutes, once
Create a token in the app, paste one command into your client. There is nothing to install and nothing to keep running.
1. Create a token
In Polyz, open Settings → Connect AI, choose what the token may do, and create it. The token is shown once. If you lose it, revoke it and make another — and revoking takes effect on the very next request.
2. Add it to Claude Code
The app gives you this command with your token already in it:
claude mcp add --transport http polyz \
https://api.polyz.ai/mcp \
--header "Authorization: Bearer YOUR_TOKEN"3. Check it works
Ask your AI to run whoami. It will tell you which Polyz account it reached and exactly what the token is allowed to do.
Codex
Codex reads the token from an environment variable rather than a header, so set it first, then add the server:
export POLYZ_TOKEN=your_token
codex mcp add polyz \
--url https://api.polyz.ai/mcp \
--bearer-token-env-var POLYZ_TOKENPut the export line in your shell profile so it survives a restart — Codex reads the variable each time it starts, and the server simply will not appear if it is unset. Check it worked with codex mcp list.
Anything else
Any client that speaks MCP over Streamable HTTP with a bearer token can connect. There is no session to maintain and no callback URL to register, so these three lines are everything yours needs:
URL https://api.polyz.ai/mcp
Transport Streamable HTTP
Header Authorization: Bearer YOUR_TOKENIt knows the book, not just the file you pasted
Your AI sees your project the way you do — as a structure it can navigate — instead of a wall of text with no shape.
Your binder, as context
It can list your projects and walk the structure of a book: folders, chapters, word counts, synopses, labels and status.
Two kinds of search
Find an exact phrase before editing it, or find the passages that are about something even when they never use those words.
@-mention a chapter
Every document is addressable, so you can point your AI at the scene you mean instead of describing it and hoping.
It proposes. You decide.
Nothing an AI writes goes into your manuscript as a finished change. Suggestions are the default, and turning that off is a deliberate choice you have to make.
Edits arrive as suggestions
Changes land as tracked insertions and deletions in your manuscript. You accept or reject each one, in the editor, at your own pace.
A snapshot before every write
Taken before the change, not after, so undo survives closing the app — even if the edit landed while you were asleep.
Editorial slash commands
/polyz:critique-chapter, /polyz:tighten-prose, /polyz:continue-scene and /polyz:continuity-check arrive with the command, already knowing how we think about editing.
How a review actually goes
- 1. You ask your AI to tighten a chapter. It reads the chapter and proposes a handful of specific changes — deliberately a handful, because a hundred suggestions is not a review.
- 2. The changes appear in your editor as you work, marked as insertions and deletions. If the document is open, it updates in place.
- 3. You accept the ones that are right and reject the rest, one at a time or all at once. Nothing expires; there is no rush.
- 4. Anything still unreviewed stays out of your exports and your word count. A suggestion is not part of your book until you say it is.
Have it teach you something
Point your AI at anything you want to learn and it will build you a real course inside Polyz — one you take with progress, quizzes and grading, not a chat transcript you will never reread.
Your AI writes the whole course
Syllabus, lessons and questions. We give it the format and a validator; the teaching is its own.
Validation is free and instant
The format check is ordinary code — no model, no cost, no waiting. Your AI iterates against it until the unit is right.
Start before it finishes
Units appear in your library as they are submitted, so you can begin unit one while your AI is still writing unit six.
Start with /polyz:build-course. It asks what you want to be able to do, agrees an outline with you, and then writes the units one at a time.
You decide what it can touch
Two independent controls, both enforced on our server rather than requested politely of the model.
What the token may do
Chosen when you create it, and changeable only by making a new one:
- Read documents — read your projects, documents and story bible.
- Write documents — draft and edit, as suggestions you review.
- Author courses — write and submit courses to your library.
A read-only token cannot be talked into writing. The permission is checked before the tool runs, not by the model.
Where it may go
Set per folder in your binder, and it applies to every AI — ours and yours:
- No access — invisible. Not readable, not searchable, not listed.
- Read only — can be read, cannot be changed. This is the default.
- Edit — existing documents can be changed.
- Full — documents can also be created.
A hidden folder does not appear in search results or listings, so your journal stays yours even when you ask for a project-wide rewrite.
Your text goes to your AI provider. When you connect your own client, the chapters it reads are sent to whoever you are paying — Anthropic, OpenAI, or whoever else — under their terms, not ours. That is the trade for running on your own subscription, and it is worth knowing before you connect a manuscript you care about.
Your AI. Your manuscript. Your call on every change.
Included in every plan. Create a token in Settings and connect in about three minutes.