Build a scene index
Instead of pasting the whole manuscript, you let mcp-writing index your scenes: where they happen, who appears, what changes, and what job each scene is doing.
mcp-writing indexes scenes, characters, beats, and revision metadata so an AI assistant can answer structural questions, trace arcs, and propose edits without loading your whole book at once.
A full manuscript is too large and too messy to paste into chat every time. The better pattern is simple: let the AI search the structure first, then read only the scenes that matter.
The workflow is simple: build a scene index first, ask structural questions second, and only load prose when the assistant needs to read or revise a specific scene.
Instead of pasting the whole manuscript, you let mcp-writing index your scenes: where they happen, who appears, what changes, and what job each scene is doing.
The AI can trace character arcs, find related scenes, compare beats, spot out-of-date metadata, and answer continuity questions before it loads any prose.
When prose is needed, the AI loads the specific scenes involved. If it suggests edits, you see the proposed change first and choose whether to commit it.
mcp-writing works beside your project instead of replacing it. Your files stay local, your structure stays visible, and every suggested prose change remains yours to approve.
Start mcp-writing beside your writing project, connect Claude Desktop, OpenClaw, or another MCP client, then run sync() so the assistant can query your scene index before it reads prose.
# Clone and install git clone https://github.com/hannasdev/mcp-writing cd mcp-writing && npm install # Point at your manuscript folder and start WRITING_SYNC_DIR=~/Documents/MyNovel/sync npm start # Connect Claude Desktop, OpenClaw, or another MCP client to: http://localhost:3000/sse http://localhost:3000/healthz # Then ask your assistant to run: sync()
Use the scene index to answer questions that normally require rereading half the manuscript: arcs, continuity, pacing, scene placement, review bundles, and safe edits.
get_arc across the project.get_scene_prose.Built for writers who want AI help with structure, continuity, and revision without handing over control of the manuscript.
Works with any folder of Markdown scene files. Scrivener External Folder Sync is supported with stable binder-ID scene identity.
Find scene titles, loglines, tags, and structural patterns quickly.
Filter by beat, character, chapter, part, POV, or tag.
Propose, review, and commit scene edits. Nothing is written without approval.
Keep scene purpose, characters, tags, beats, and revision notes attached to the manuscript.
When prose changes, the system can warn that scene metadata needs refreshing.
Structured character and place sheets keep world-building consistent across the project.
Export scoped PDF or Markdown packages for editors and beta readers.