Slima MCP

Your AI tools,
reading and writing your book.

Through the Model Context Protocol, Claude, Cursor, Gemini CLI, and VS Code reach into your Slima books directly. One cloud, many AI clients, no copy-pasting.

Get started See full workflow
  • Claude · Cursor · Gemini · VS Code
  • Read / write / search / analyze
  • OAuth · no stored passwords
Claude Desktop
Connected · slima-mcp
U
Pull up Chapter 8 of Eleanor and the Boats. Tighten Maya's dialogue — sharper, fewer beats.
→ slima_get_chapter(book="...", number=8)
→ slima_edit_file(path="ch08.md", changes=...)
★
Read chapter 8. Tightened Maya's six lines, sharper edges. Written back to Slima.
— Open standard

MCP is Anthropic's open standard for AI integrations.

Anthropic introduced the Model Context Protocol in late 2024 to let AI assistants safely connect to outside tools and data. Slima is one implementation — your work stays in Slima's cloud, and AI clients reach in only with the access you grant.

— 01 · Workflow overview

From AI tool to your book, in four steps.

Each step solves a place where integrations usually break — and the whole flow takes under five minutes.

Step 01
Connect

Connect

Pick your AI tool

Claude Desktop, Cursor, Gemini CLI, VS Code, any MCP-compatible client — they all connect to the same Slima account.
Step 02
Auth

Authenticate

OAuth, no stored passwords

One command opens your browser to sign in. The client gets an access token — you can revoke it anytime, and you always see who's connected.
Step 03
Read · Write

Read & write

The whole book in one tool set

AI can list books, read chapters, edit files, create notes, search across the whole manuscript, call Beta Reader, pull writing stats — all through structured tool calls.
Step 04
Multiplex

Multiplex

One cloud, many clients

Edit technical docs in Cursor, draft fiction in Claude Desktop, run research in Gemini CLI — same book, same cloud, same version history.
— 02 · Connect

Whichever AI tool you already use.

Slima MCP is built on the standard MCP spec, so any compliant client works. Below are the ones we've tested and have setup guides for:

i

Claude Desktop

Anthropic's official desktop app. Most complete MCP support, most direct UX. The default for long-form writing.

Recommended for fiction
ii

Cursor

VS Code fork with built-in AI — great for calling Slima from inside the editor. Common for screenwriting, technical writing, and doc revisions.

Recommended for scripts / docs
iii

Gemini CLI

Google's command-line AI. MCP support is stable. Right for research mode and batch pipelines.

Recommended for research / batch
iv

VS Code

GitHub Copilot Chat supports MCP too. Familiar for devs who edit prose alongside code.

Familiar to developers
v

Any MCP client

Continue.dev, Zed, your own MCP client — anything that follows the spec works. Slima MCP is an open, public protocol.

Open protocol, no lock-in
— 03 · Authenticate

OAuth flow, no stored passwords, revocable anytime.

One command starts it. Slima uses OAuth 2.0 — your password never leaves the browser, and the client gets a time-bound access token.

Run in your terminal
$ npx @slima/mcp auth 
✓ Opening browser to sign in...
✓ Authorized. Token saved to ~/.slima/mcp.json
✓ Connected clients: claude-desktop, cursor

Standard OAuth

Sign in via the browser; the client gets a token. No pasting passwords into config files, no API key sprawl.

Revocable anytime

In Slima account settings, see every connected client, when it last ran, and revoke any of them — or all of them — with one click.

Fine-grained scopes

Pick read-only or write, scope by book, set per-day call limits. AI can only do what you authorized.

— 04 · Read + Write

A structured tool set for the whole book.

AI clients call through MCP tools. Each tool is a structured function with a schema — AI can't mangle formats or accidentally delete files.

Book management

  • list_books

    List all books with metadata

  • create_book

    Create a new book (novel or script)

  • get_book_structure

    Tree view of chapters / scenes / files

  • get_writing_stats

    Word counts, streaks, progress

File operations

  • read_file

    Read a single chapter or note

  • edit_file

    Diff-based edit (no overwrite)

  • create_file

    Add chapter / character / scene file

  • search_content

    Whole-book search (with regex)

AI analysis

  • analyze_chapter

    Call Beta Reader for chapter feedback

  • list_personas

    List available reader personas

— 05 · Multiplex

One book, many AI clients.

The same book can be open in Claude Desktop, Cursor, and Gemini CLI at once. Slima is the single source of truth — version history, conflict detection, cloud sync all live in Slima.

MCP
Your book (Slima cloud)
Claude Desktop
Drafting at night
Cursor
Scene revisions in the day
Gemini CLI
Weekend analysis runs

AI clients don't talk to each other directly — every change goes through Slima cloud, every version is yours.

— 06 · How writers actually use it

Real workflows, not demo gimmicks.

01

Writing assistant mode

"Tighten Maya's dialogue in chapter 8, sharper edges." Claude reads the chapter, edits, writes back. You don't switch windows.

02

Lookup mode

"Where does Maya say she speaks three languages?" AI searches the whole book, returns the quote with chapter location.

03

Research pipeline

Gemini CLI batch-runs Beta Reader across 30 chapters, returns 30 reports, compiles them into a master spreadsheet.

04

Cross-device drafting

Cursor by day, Claude Desktop at night, Slima web on the train — same book, no version chaos.

— 07 · Included on every plan

MCP is a free Slima feature.

Free works. You pay your AI client (Claude / Cursor subscription) for compute on their side — Slima doesn't charge anything for the protocol itself.

Slima Free

$0

  • Full MCP support
  • All tools available
  • OAuth authentication
  • Up to 2 books at a time
Start free

Slima Starter / Pro

Unlocks usage limits

  • Everything in MCP
  • Unlimited books (Starter+)
  • Beta Reader report quotas (per plan)
  • Higher daily MCP call limits
See plans
— 08 · Common questions

"Is my data safe? Will AI mangle my draft?"

If your question isn't here, write to us. A human reads everything.

Will AI vendors (Anthropic, OpenAI, etc.) train on my book once they see it? +

Slima never trains on your work — that's written into our terms. Whether the AI client trains depends on your subscription with them: Claude defaults to no training, Cursor's business plan doesn't train, others vary. MCP is a transport protocol — it doesn't decide training policy.

If AI edits something wrong, what do I do? +

edit_file is diff-based. Every change goes into Slima's version history — you can revert any version with one click. AI never overwrites the whole file; each edit is atomic.

Can I limit AI to read-only? +

Yes. In Slima account settings, set permissions per client: read-only / write / full, scope by book, set daily call limits. A common setup is Claude Desktop read-only and Cursor with write access.

MCP is Anthropic's standard — does ChatGPT support it? +

MCP is open. Anthropic proposed it but any client can implement. ChatGPT desktop doesn't have native MCP yet — bridge tools exist (community-maintained). Cursor, Continue.dev, and Zed are built-in.

Why not just use ChatGPT GPTs or Claude Projects? +

GPTs / Projects work for pasting one document and asking questions. MCP is "persistent, two-way, structured" — AI reads the latest version on demand, can write back, and can call structured tools. For long-form, the difference is large.

Does it cost extra? +

Slima MCP is free on every Slima plan. You pay your AI client (Claude / Cursor / etc.) for their compute. Slima only charges if you exceed the free book limit or want Beta Reader report quotas.

Bring AI into your manuscript.

MCP is free. OAuth authorization in one command, full tool set, version history that protects your draft.

Get started 📖 Read the docs
Free entirely OAuth authorized Revocable anytime