@getnarro/mcp-server

MCP (Model Context Protocol) server for AI-assisted Narro presentation creation

getnarro.com · Documentation · Ecosystem

The tools act on the filesystem: they scaffold projects, rewrite deck files, and run builds. Point the server at a directory you are willing to have written to.

Already using a coding agent with a terminal? You probably do not need this. Claude Code, Cursor's agent, and anything else with shell access can use the narro CLI directly — what they need is the documentation, and https://getnarro.com/llms.txt (or npm i -D @getnarro/docs) gives them that. This server is for assistants that cannot run commands themselves.

Setup

Claude Code

claude mcp add narro -- npx -y @getnarro/mcp-server

Claude Desktop

~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "narro": {
      "command": "npx",
      "args": ["-y", "@getnarro/mcp-server"]
    }
  }
}

Cursor, Cline, Continue

The same object, under whatever key the client uses for MCP servers — for Cline that is cline.mcpServers in VS Code settings.

Then ask: "Create a Narro deck about our Q3 results and make sure it builds."

Tools

| Tool | What it does | | --- | --- | | create_presentation | Scaffolds a project, returns the path to the deck file | | list_slides | Structured slide list: index, id, title, layout, preview | | add_slide | Inserts a slide and writes the file | | update_slide | Rewrites one slide, preserving frontmatter it was not asked to change | | delete_slide | Removes a slide, returning its content so it can be restored | | apply_theme | Sets the deck's theme | | check_presentation | Validates layout, slot, and theme names without building — fast enough to run after each edit | | build_presentation | Builds, and returns the compiler output on failure | | export_presentation | PPTX, PDF, Google Slides, or a static site | | bundle_presentation | One self-contained HTML file | | import_presentation | From PowerPoint or Google Slides | | get_presentation_info | Documentation, and per-component prop types | | manage_comments | Slide comments |

Video

The same loop for @getnarro/video, whose timeline is derived from the narration rather than authored — so every step below is an edit to sentences, and an assistant can run all of it. Each tool is narro-video <command> in a project directory, and each returns the command's output and the files it wrote.

| Tool | Does | | --- | --- | | video_script | The brief a model answers with the first draft of the narration, then that answer as script.ts | | video_render | The masters, reusing every scene whose inputs have not changed | | video_check | Text that leaves the frame, every citation against its source, the script against the brand policy | | video_review | The director's pass: frozen holds, dead air, pace, legibility, safe zones, late entrances, an empty poster | | video_apply | A feedback export — the studio's, or a review — as a patch to script.ts | | video_brief | The brief for a translation or a channel cut | | video_cover | The brief for a cover image, and the image from the answer | | video_retention | A platform's retention export mapped onto the sentences that cost the most audience | | video_record | Scripted product footage, captured on virtual time; check says whether the product changed | | video_stale | Which cited sources have moved since the last render | | video_approve | The one document a person approves |

Terminal demos

@getnarro/terminal holds a terminal session as text. These five are narro-terminal <command> against a playbook file, and each returns the command's output.

| Tool | Does | | --- | --- | | terminal_render | The playbook as an animated GIF, or an SVG with svg | | terminal_still | One frame as a PNG — at a mark, a second or a frame | | terminal_check | What will be wrong before anything renders: unimplemented sequences, lines wider than the grid, a mask that matches nothing, timing that will read badly | | terminal_import | An asciicast, a VHS .tape or a ttyrec, read into a playbook | | terminal_export | A playbook written back out as an asciicast |

There is no terminal_record tool, deliberately. Every tool above draws text; record runs the playbook's commands in a real shell, and the CLI guards it by printing the commands and asking a person. A tool would route around exactly that gate, so an assistant authors, renders and checks a demo — and a person runs it. Nor is there a terminal_tape: narro-terminal tape opens a shell for a person to perform the demo in, and there is no person at the other end of a tool call. What it writes — a tape, a playbook and a capture — every tool above reads.

Recordings

@getnarro/replay records a web app as real DOM. Recording is a person's job — it drives a browser through a flow — so there is one tool, and it reads.

| Tool | Does | | --- | --- | | replay_info | A cast's viewport, duration, size and chapter names | | replay_review | What is wrong with the recording, and what to do about each one |

Call replay_info before writing <Replay chapter="…"> or <narro-replay start="…">. Those name a chapter as a string, and a name the cast does not have is not an error: seek falls back to the beginning, so the demo plays from the top and reads as a timing bug rather than a typo.

Call replay_review before publishing one. A cast is text that travels, and an email address or an API token typed into a form during the recording is in that text. The rest of the pass — a click that changed nothing, dead air, a scroll nobody could read, type that will be six pixels tall on a page — is about whether the demo is worth showing.

There is no replay_apply, deliberately. video_apply exists because rewriting narration in a video moves every cue after it, so the patch has to be computed. A recording has no such coupling: a finding asks either for a re-record, which is a person driving a browser, or for an edit to the .notes.md track — an ordinary markdown file an assistant edits directly and bakes in with narro annotate.

build_presentation is the one to remember. A generated deck that looks right can still be broken; without a pass/fail signal there is nothing to correct from.

Editing decks

add_slide, update_slide, and delete_slide edit markdown decks. Slides are addressed by id where one exists, because an index shifts the moment something is inserted above it — and a tool that silently edits the wrong slide is worse than one that fails.

React decks are read-only to these tools: list_slides works, but slides there are React components and the file should be edited directly.

Resources

The documentation is exposed as resources at narro://docs/{slug}, which a client can read and cache without spending a tool call. get_presentation_info serves the same pages for clients that do not support resources.

Both read @getnarro/docs, whose component reference is generated from the TypeScript source — so prop types describe the shipping API rather than a hand-maintained copy of it.

The Narro ecosystem

| Package | What it is | | --- | --- | | @getnarro/annotate | Feedback on a Narro recording — typed notes anchored to a moment, staleness against the cast they were written about, and a briefing a coding agent can act on | | @getnarro/atlas | A spatial canvas for Narro — a camera path as text, one pure camera function over it, and scene providers a deck, a video and a web page all render the same way | | @getnarro/cli | CLI tools for creating and managing Narro presentations | | @getnarro/core | The React runtime for Narro presentations — Presentation, Slide, navigation, transitions, and speaker notes | | @getnarro/docs | Narro's documentation as data — markdown pages, a generated component API reference, and llms.txt | | @getnarro/docspack | Narro's documentation as a docspack — version-locked chunks an agent installs once and searches offline | | @getnarro/markdown | Markdown & MDX authoring for Narro presentations | | @getnarro/marketplace | Themes, colour schemes, and templates for Narro presentations | | @getnarro/mcp-server | MCP (Model Context Protocol) server for AI-assisted Narro presentation creation ← you are here | | @getnarro/replay | Record a web app once and replay it as real DOM — a portable cast format, an embeddable player, and a self-contained HTML page | | @getnarro/shared-ui | Slide components for Narro presentations — headings, text, lists, code, media, charts, and layouts | | @getnarro/terminal | Terminal playbooks for Narro — a text format for a terminal session, a screen that is a pure function of the frame number, and a renderer that draws it on a grid | | @getnarro/video | Frame-deterministic video framework for Narro — a brand system, a React timeline runtime, narration-derived timing, a headless-Chrome renderer, and a scrub studio | | @getnarro/video-marketing | Marketing scene components for @getnarro/video — lower thirds, stat counters, feature grids, pull quotes, device frames, callouts and end cards, all brand-driven and frame-deterministic | | create-narro | Scaffold a new Narro presentation |

All fifteen ship from one repository and release together.

The npm package named narro is unrelated to this project. Narro's packages are all under the @getnarro/ scope; the CLI binary narro comes from @getnarro/cli.

Website · Documentation · llms.txt · GitHub · Issues · Changelog

Released under the MIT License.