For AI assistants and developers

MechaSquid is a writing app for novelists: a quiet place to write the book, and a living story bible (characters, places, timeline, outline, continuity checks) that the author's own AI assistant, such as Claude or ChatGPT, can read and keep up to date through MCP. The prose stays the author's.

Connect

• Claude, ChatGPT and other apps that sign in: add the address as a custom connector. The app registers itself and signs in with OAuth 2.1 (PKCE); the author sees which app is asking and allows it on mechasquid.com. One connection reaches all of the author's worlds; tools take an optional world argument.
• Your own agent: the author creates an API key in a book's Settings. Send it as an X-Api-Key header, or as a Bearer token, to the same address. A key reaches one world only.
• First call: help with no topic returns the overview below and the list of topics. Every refusal names the tool to call next.

Resources you can attach

Besides its tools, MechaSquid offers context you attach to a conversation yourself (in Claude: the + menu, Add from MechaSquid; in Claude Code: an @ mention). Nothing is read until you attach it.

• Guides: how MechaSquid works (Using MechaSquid) and each craft guide.
• Per world: each book's outline with its plans, the writing brief (your voice), the cast, the timeline and the open flags.
• A scene with its context: chosen by name; the scene with its characters, places, what surrounds it, open flags and comments.

Addresses look like mechasquid://world/<id>/cast. Not every app shows resources yet; the tools work either way.

The rules every agent works under

• The author's prose changes only with their yes: an agent proposes a revision and it waits for the author, who accepts it on the website or says yes in the conversation. Direct edits happen only where the author switched them on.
• Linking names never changes a visible word; the server checks and refuses otherwise.
• A book marked final is read-only to agents.
• Everything an agent does is shown to the author under the app's name, and they can undo it or disconnect the app at once.

Guides

The same text the help tool serves to connected agents.

• Using MechaSquid

  Start here. What MechaSquid is, the rules every agent works under, and which tool or help topic fits what the author just asked.

• Capturing ideas

  For quick sessions, often by voice on a phone. Save the thought fast, in the right book, without questions the author cannot answer on a walk.

• Continuity and health

  Timeline, scene context, the health check and flags; how to find and report contradictions.

• Process newly written text

  File what the author wrote in Write mode (scenes, entities, links, consistency, copyedit) without changing their words.

• Maps and travel

  For where places lie and how long a journey takes. How to record what the author says about travel, and how to ask MechaSquid for a route before writing one.

• Non-fiction books

  Read when the book is not a novel: self-help, a how-to, memoir, essays, popular science. How MechaSquid's novel-shaped pages map onto it, and what to leave alone.

• Plotting and pacing

  Fetch before talking about the shape of a book: its high and low points, its threads, planning scenes that are not written yet. Offers shapes; imposes none.

• Revisions and comments

  How to propose changes to the author's prose, answer their notes, and what happens when they accept or reject.

• Books, pages and links

  How a book is organised (books, chapters, scenes, wiki pages, slugs, links, relationships) and the tools that read and change it.

• Writing in the author's voice

  The writing brief, character voices, and how to turn the author's rejections into rules they can accept.

• Building the world

  For conversations about the world itself (magic, history, politics, a character's past), with or without a book yet. How to turn talk into wiki pages the author can trust.

Support

Questions, problems, or anything about your data: support@mechasquid.com. A connected agent can also report a problem itself with send_feedback. Privacy: privacy policy.

A plain-text summary for language models is at /llms.txt.