Built for novelists working with long manuscripts that don't fit in a single context window. Syncs from Scrivener or native markdown, builds a SQLite index of scenes, characters, places, and relationships, then lets the AI query that index before loading prose. Core operations: sync project structure, find scenes by tag or character, get character arcs across chapters, read specific scene prose, edit with confirmation and git history, flag continuity issues. Includes backup and restore tooling, sidecar metadata maintenance, and review bundle generation. Designed to pair with OpenClaw but works with any MCP client over stdio. Ships with agent boot files and workflow guidance for continuity passes, subplot tracking, and targeted revisions without feeding the full manuscript into every prompt.
claude mcp add --transport stdio hannasdev-mcp-writing uvx mcp-writing