scrivener-mcp

Scrivener MCP

The definitive MCP server for Scrivener — connect your novels, screenplays, and manuscripts to Claude, ChatGPT, and any AI assistant. 53 tools: document management, writing analysis, content enhancement, offline semantic search, and character/plot tracking.

·

·

·

·

Scrivener MCP lets your AI assistant open, read, edit, analyze, and search your Scrivener projects directly. No copy-pasting. No exporting. Tell your assistant which project to open, and start working.

> You: Open my novel and analyze the pacing in Chapter 12.

>

> Claude: *Opens your .scriv project, reads Chapter 12, runs pacing analysis.*

> The first half moves well with short, tense paragraphs. The middle section slows

> considerably -- the three-page internal monologue starting at paragraph 14 stalls

> the momentum you built in the confrontation scene. Consider cutting it to a single

> paragraph and moving the backstory to Chapter 8 where Elena is first introduced.

Works with Claude Desktop, Claude Code, VS Code (Copilot/Continue), Cursor, and any MCP-compatible client. Scrivener 3 on macOS, Windows, and Linux. Listed on the official MCP Registry as `io.github.writerslogic/scrivener-mcp`.

Install

Pick the method that works for you. Most auto-configure Claude Desktop on install. Claude Code and other clients need one extra step -- see Claude Code below.

npm (recommended)

npm install -g scrivener-mcp

Restart Claude Desktop. Done.

Claude Code

Installing the npm package does not register the server with Claude Code -- the install-time auto-config only writes Claude Desktop's config. After installing, register the server:

npx scrivener-setup

This detects Claude Code (along with Claude Desktop and Cursor) and writes the config for you. To register it manually instead:

claude mcp add -s user scrivener -- npx scrivener-mcp

Then restart Claude Code (or run `/mcp` to reconnect) and Scrivener MCP appears in the server list. Drop `-s user` to scope it to the current project instead of all projects.

Smithery

npx -y @smithery/cli install scrivener-mcp --client claude

npx (no install)

Use directly without installing globally:

npx scrivener-mcp

Or add to your Claude Desktop config manually:

{
  "mcpServers": {
    "scrivener": {
      "command": "npx",
      "args": ["scrivener-mcp"]
    }
  }
}

GitHub

Install directly from the repo (latest main):

npm install -g writerslogic/scrivener-mcp

Or a specific release:

npm install -g writerslogic/scrivener-mcp#v0.12.0

Homebrew (macOS)

brew install writerslogic/tap/scrivener-mcp

Docker

docker build -t scrivener-mcp https://github.com/writerslogic/scrivener-mcp.git
docker run -i --rm -v /path/to/your/projects:/projects scrivener-mcp

Setup for other MCP clients

Run the interactive setup to auto-detect and configure your client:

npx scrivener-setup

This detects Claude Desktop, Claude Code, and Cursor, and writes the config for you.

For other MCP clients, point them at `npx scrivener-mcp` as a stdio server.

Optional: AI-powered features

Core features (document management, deterministic analysis, keyword search, and project memory) work without any API key. AI-powered analysis, generation, enhancement, and semantic search work with an Anthropic (Claude), OpenAI, or OpenRouter key; when several are present, Claude handles chat and generation (set `AI_PROVIDER=openai` or `AI_PROVIDER=openrouter` to override). OpenRouter defaults to the `anthropic/claude-sonnet-4.6` model; set `OPENROUTER_MODEL` to use another model in its catalog. If the active provider fails with an account-level error (invalid key, exhausted credit, outage), the server automatically retries the request on the next configured provider. When your MCP client supports the sampling capability, supported chat-based AI features can also run through the client's own model—with no separately configured API key. Semantic indexing and similarity scoring use the local Holographic Memory System rather than an external embedding API, while the current `semantic_search` pipeline uses the configured chat provider to interpret queries and explain results. The server automatically discovers keys from common locations:

• `ANTHROPIC_API_KEY` / `OPENAI_API_KEY` / `OPENROUTER_API_KEY` environment variables
• `~/.env`, `~/.scrivener-mcp/.env`
• `~/.anthropic/key`, `~/.openai/key`, `~/.openrouter/key`
• macOS Keychain (service names `anthropic-api-key` / `openai-api-key` / `openrouter-api-key`)

To store a key in the macOS Keychain:

security add-generic-password -s anthropic-api-key -a anthropic -w sk-ant-your-key-here

Or export it manually:

export ANTHROPIC_API_KEY="sk-ant-..."   # or OPENAI_API_KEY="sk-..."

This enables provider-backed writing analysis, content enhancement, generation, semantic search, character consistency checking, and intelligent compilation.

What You Can Do

> First, open a project. The server acts on whatever `.scriv` project you point it at -- it has no link to the Scrivener app and can't see what you have open there. Start a conversation with *"Open my Scrivener project at `~/Documents/My Novel.scriv`"* (or *"Discover my Scrivener projects"* if you don't know the path), then give your commands. On macOS you can also just say *"Use the project I have open in Scrivener"* -- it detects the open project and opens it (the first time, macOS asks you to allow controlling Scrivener). Do this once at the start of each conversation; the examples below all assume a project is open. If the same project is also open and unsaved in the Scrivener app, save or close it there first to avoid conflicting writes.

Manage Your Manuscript

Open any Scrivener project and work with it naturally. Read chapters, create new scenes, reorganize the binder, update synopses -- all through conversation.

> You: Create a new scene called "The Reveal" after Chapter 5, and move the old epilogue to the trash.

Analyze Your Writing

Get detailed feedback on readability, pacing, style, dialogue quality, and emotional arc. Not generic advice -- analysis grounded in your actual prose.

> You: Analyze Chapter 3. Is the pacing too slow?

>

> Claude: Readability is good (Flesch-Kincaid grade 8.2), but pacing flags:

> - 4 consecutive paragraphs of internal monologue (lines 45-78) with no action or dialogue

> - The scene is 3,200 words with only 2 scene breaks -- your other chapters average 4

> - Filter word density is 2x your manuscript average ("felt", "seemed", "noticed")

> Specific suggestions: ...

Enhance Your Prose

Apply targeted improvements: eliminate filter words, strengthen verbs, vary sentence structure, add sensory details, convert telling to showing, tighten dialogue, adjust pacing.

> You: Eliminate the filter words in Chapter 7 and strengthen the verbs.

Track Characters and Plot

Store character profiles, plot threads, and style guides that persist with your project. The AI remembers your characters across sessions.

> You: Save a character profile for Marcus: retired detective, cynical but fair, walks with a limp from an old injury, speaks in clipped sentences.

>

> *Later...*

>

> You: Check if Marcus is consistent across all chapters.

>

> Claude: Found an inconsistency: Marcus walks "briskly" in Chapter 9 (line 34), but his limp is referenced in Chapters 2, 5, and 11. Also, his dialogue in Chapter 4 uses long flowing sentences, which contradicts the "clipped sentences" note in his profile.

Search by Meaning

Find passages by what they're about, not just keyword matching. "Find scenes where the protagonist feels isolated" works even if the word "isolated" never appears. The project index and similarity scoring run locally through the Holographic Memory System; the current search pipeline also uses your configured AI provider for query interpretation and result explanations, so `semantic_search` requires a provider.

> You: Find all scenes where Elena and Marcus are alone together.

Track Relationships

Store and query relationships between characters, locations, themes, and plot threads. No Neo4j required -- relationships live in the semantic memory engine and persist with your project.

> You: Who is connected to Marcus? What plot threads involve the lighthouse?

Compile and Export

Combine chapters into a single manuscript with configurable formatting, separators, and structure preservation. Export the result inline as Markdown, HTML, or JSON, or write a DOCX, EPUB, or PDF file to disk for submission, e-readers, or print.

All Tools

57 tools organized by workflow. To keep token usage low, tools load progressively -- project tools at startup, document and search tools when you open a project, and the rest on demand (your AI client activates them automatically, or calls them directly and the owning skill activates on the fly). Set `SCRIVENER_MCP_EAGER_TOOLS=1` to load everything at once.

Project -- open, browse, manage

Documents -- read, write, create, organize

Search -- find content, passages, and mentions

Analysis -- quality, consistency, structure

Enhancement types: `eliminate-filter-words`, `strengthen-verbs`, `vary-sentences`, `add-sensory-details`, `show-dont-tell`, `improve-flow`, `enhance-descriptions`, `strengthen-dialogue`, `fix-pacing`, `expand`, `condense`, `rewrite`

Compile & Export -- assemble and ship the manuscript

Memory -- persistent project knowledge

Memory is stored within each .scriv project and travels with it.

Relationships -- entity connections and story graph

Works without Neo4j -- relationships live in the Holographic Memory System and are available immediately. The document cross-reference tools are fully deterministic (exact whole-word matching, no AI) and need no external services; Neo4j adds advanced graph analysis when connected.

Background Jobs -- long-running analysis

Discovery -- explore capabilities

Guides

• **Getting Started** -- Installation, configuration, your first session
• **MCP Client Setup** -- Copy-paste config for Claude Desktop, Claude Code, Cursor, and VS Code
• **Writing with AI** -- Analysis workflows, enhancement strategies, memory management
• **Troubleshooting** -- Common issues and fixes
• **Token Optimization** -- How the server minimizes context window usage
• **Architecture** -- How the server works, module structure, data flow
• **Scrivener Compatibility** -- Supported Scrivener versions, platforms, and format coverage
• **Scrivener File Format** -- The reverse-engineered `.scriv` format, what we read vs. infer, and safe-modification guidance
• **Fuzzing** -- Jazzer.js target and OSS-Fuzz integration details
• **Contributing** -- Development setup, code conventions, adding new tools

Requirements

• Node.js 18+
• Scrivener 3 project files (.scriv)
• macOS, Windows, or Linux
• Optional: Anthropic, OpenAI, or OpenRouter API key for provider-backed AI features
• Optional: Neo4j for persistence and advanced graph queries; core relationship tools work without it

Development

git clone https://github.com/writerslogic/scrivener-mcp.git
cd scrivener-mcp
npm install
npm run dev          # Development mode with hot reload
npm run build        # Compile TypeScript
npm test             # Run tests
npm run typecheck    # Type checking only

Why This One?

Several Scrivener MCP servers exist. Feature claims below come from each project’s public documentation, published package, and advertised tool surface, last re-read on 2026-08-22; stars, forks, activity, and published version were refreshed 2026-08-22. “No” means the project does not document that capability; it does not claim the capability is impossible through the connected AI client.

Counts and feature claims can change. Follow the linked projects for their latest documentation.

The table is generated from `docs/comparison.yml` — edit claims there, not here.

The option that isn't an MCP server

Worth naming, because it is the real alternative for many writers: Scrivener can

Sync to External Folder, writing each document out as RTF or plain text, and any

general-purpose file MCP server (for example

`@modelcontextprotocol/server-filesystem`)

can then read and write those files.

That costs nothing and works today. What it gives up is everything that depends on

understanding the project rather than the folder: the binder hierarchy, metadata,

labels and status, snapshots, compile settings, and RTF formatting all flatten away,

and edits land in the sync folder rather than the project — so a bad edit is

reconciled by Scrivener on the next sync rather than caught before it happens. Use the

sync-folder route for occasional read-only help with prose; use a Scrivener MCP server

when you want the structure to survive the round trip.

Contributing

We welcome contributions of all sizes. Check the issue tracker for `good first issue` labels, or see the contributing guide for development setup.

Areas where help is especially welcome:

• Test coverage (#18)
• Windows testing and path handling
• Scrivener 2 compatibility testing
• Documentation improvements (#25)

Security

Found a vulnerability? Please report it privately — see SECURITY.md.

License

AGPL-3.0 © WritersLogic, Inc.

Free for personal use and open-source projects. Commercial license available for proprietary integration. See COMMERCIAL_LICENSE.md for details.

·

·

·