Copy. Paste. Copy again. Paste again — drop your character sheet into the chat box, then paste chapter eight, just to ask the AI one question: “do these two contradict each other?” Spend an afternoon and you’ve moved more text than you’ve written.
The problem is simple: the AI can’t see your whole book. It only sees the fragments you hand-feed it. Slima MCP fixes exactly that — set it up once, and your AI can read and write your Slima books directly. No more copy-paste.
This guide starts with how to install it, walks through every method, then shows you what to actually do with it.
What is MCP? Short for Model Context Protocol, it’s an open standard that lets AI tools exchange data with outside services directly. Think of it as USB for AI: any tool that speaks MCP can plug into Slima and read or write your book.
Pick your path
| How you use AI | Which install |
|---|---|
| claude.ai or ChatGPT in a browser / on your phone | ⭐ Web connector (one-time setup, auto-updates forever, no token) |
| Claude Desktop / Cursor / Gemini CLI on a computer | Local npx install |
| Enterprise / air-gapped | Self-host (advanced) |
The first one is the easiest — especially if you ever write on your phone.
Method 1 (recommended): connect from claude.ai or ChatGPT on the web
If you chat with Claude or ChatGPT in a browser or a phone app, this is the simplest and most future-proof option. Set it up once and every Slima MCP update lands automatically — no commands, no config files.
All you need is one URL: https://mcp.slima.ai/mcp
Connect from claude.ai (my own go-to)
- Sign in at claude.ai.
- In the left sidebar, open Customize → Connectors.
- At the top right of the connector list, click ”+” → Add custom connector.
- Fill in just two fields:
- Name:
Slima - Remote MCP server URL:
https://mcp.slima.ai/mcp - Leave the OAuth Client ID / Secret blank.
- Name:
- Click Add → you’re sent to slima.ai to approve → you return to claude.ai.
- The Slima tools icon appears next to your chat box. Done.
The web connector is available on the claude.ai Pro / Max / Team / Enterprise plans.
Connect from ChatGPT on the web
- Sign in at chatgpt.com.
- Open Settings → Connectors → Advanced and turn on Developer Mode (2025.10+).
- Back in Connectors, click Create → Custom MCP Server.
- Fill in:
- Name:
Slima - Server URL:
https://mcp.slima.ai/mcp - Authentication: OAuth
- Name:
- Click Create and finish the OAuth approval.
After that, Slima’s tools show up automatically on your next conversation. No restarts, no reinstalls.
Method 2: local install with npx (Claude Desktop / Cursor / Gemini CLI / VS Code)
Use this if you prefer desktop apps, want faster startup, or run MCP alongside other local servers. The core is the same everywhere: have the tool run slima-mcp@0 with npx.
Authenticate once (saves your login locally):
npx slima-mcp@0 auth
A browser opens and walks you through signing in with your Slima account; it’s saved to ~/.slima/credentials.json. Or grab a token from Slima Settings → API Tokens and put it in the config’s SLIMA_API_TOKEN.
Claude Desktop
Edit claude_desktop_config.json (~/Library/Application Support/Claude/ on macOS, %APPDATA%\Claude\ on Windows):
{
"mcpServers": {
"slima": {
"command": "npx",
"args": ["-y", "slima-mcp@0"],
"env": { "SLIMA_API_TOKEN": "slima_your_token_here" }
}
}
}
Restart Claude Desktop; the tools icon next to the chat box means you’re connected. (If you already ran auth, you can drop the env block.)
Cursor
Create .cursor/mcp.json in your project root with the same content as Claude Desktop above. Cursor detects it and starts it on demand.
Gemini CLI
One line:
gemini mcp add slima --command "npx -y slima-mcp@0"
VS Code
Create .vscode/mcp.json:
{
"servers": {
"slima": {
"command": "npx",
"args": ["-y", "slima-mcp@0"]
}
}
}
⚠️ VS Code uses servers, not mcpServers — the single most common mistake.
Why
slima-mcp@0instead ofslima-mcp?@0pins to the current major version: npx still pulls new0.xreleases automatically (so you get features and fixes), but when a breaking1.0ships it won’t silently update under you.Prefer a global install?
npm install -g slima-mcpworks too — just runnpm install -g slima-mcp@latestnow and then so your local copy doesn’t drift from the tools Slima’s backend exposes.
Auth & security
npx slima-mcp@0 auth— sign in via browser (recommended)npx slima-mcp@0 status— check your auth statusnpx slima-mcp@0 logout— remove saved credentials
The web connector uses OAuth — no tokens copy-pasted — and you can revoke access anytime from Slima Settings. Local tokens live in ~/.slima/credentials.json with protected permissions. Slima MCP only ever touches the books in the account you authorized, and the AI only reads or writes when you ask it to.
Fifteen tools, three groups
Book management
create_book new book · list_books list everything · get_book book details · get_book_structure the whole file tree (gives the AI the big picture, so every later move is sharper) · get_writing_stats writing stats.
File operations
read_file / get_chapter read a file (the workhorse) · write_file overwrite a file · edit_file precise search & replace · create_file new file · delete_file delete · append_to_file add to the end · search_content full-text search across every file (the killer tool for consistency).
AI Beta Reader
list_personas list the reader personas · analyze_chapter have a chosen persona read a whole chapter.
Note: these all work on Writing Studio books. For Script Studio (screenplays), the structured files (scenes, characters, storylines…) are read-only over MCP — edit those in the Slima app.
Real workflows
1. Start a writing day
“Show me this book’s structure, then read the last chapter I wrote yesterday.” The AI runs get_book_structure + read_file, hands you the context, and tells you where you left off and what the outline says comes next.
2. Character consistency check
“Search every passage where the old butler appears.” search_content pulls them up — and you spot that he swore he never lies in chapter three but is clearly hiding something in chapter ten. Planted foreshadowing or a slip, at least now you caught it.
3. Revision (let the Beta Reader read)
“Read chapter one as the pacing analyst.” analyze_chapter gives you feedback on pace, tension, and character arc; then “trim that overlong scene in the third paragraph” and edit_file changes just that, leaving everything else alone.
4. Worldbuilding cleanup
“Search every passage that mentions a magic-system rule and compile them into one reference file in the worldbuilding folder.” search_content → read_file → create_file turns scattered fragments into a single source of truth.
5. Research-heavy fiction (field research)
Historical, regional, or trade-specific stories pile up research notes, interview transcripts, and location files. Drop them into a Slima folder and ask: “Compare my courtroom scene in chapter six against ‘research/court-procedure.md’ — does anything not line up?” Or “Pull the dock-work details from my interview notes into a scene reference and save it to the folder.” The AI reads your research files directly, cross-checks across them, and folds the facts back into the book — no more relying on memory.
6. On the go, collaborate from your phone
Because the web connector is cloud-based with no local setup, you can use it from the claude.ai app on your phone. On the train, in line, at a café — when an idea hits, just say “Read the last scene of chapter eight and give me three ways it could go next,” or “Add this line of dialogue to the end of ‘ideas.md’.” Open Slima back at your desk and it’s all synced. Your phone captures and brainstorms; your desktop does the real writing — and nothing gets emailed between them.
FAQ
Can anyone else see my content? No. MCP only touches the books in the account you authorized; the AI reads or writes only when you ask, and never shares anything.
Can I let the AI read but not edit? Yes. Just say “read-only today” up front and it’ll stick to the read tools (read_file, search_content).
Tools not responding after setup? For local installs it’s almost always a JSON typo (extra comma, missing brace, full-width quotes) — validate the file, then run npx slima-mcp@0 status. For the web connector, double-check the URL is https://mcp.slima.ai/mcp and that you finished the approval.
Do I need Node.js on my phone? No. On mobile you use the web connector (the claude.ai / ChatGPT app) — no command line at all.
Three minutes, a permanently better workflow
The shortest path: in claude.ai, go to Customize → Connectors, add a custom connector with the URL https://mcp.slima.ai/mcp, approve it, done. From then on — desktop or phone — just tell the AI “show me where I left off yesterday” and it goes and reads your book itself.
Time should go into the story. Not into copy-paste.