HTML/CSS/JS to video renderer with deterministic animation capture and TTS narration for LLM agent pipelines
claude-explains is a specialized HTML-to-video rendering toolkit with 17 tools. While tool names follow verb_noun conventions (analyzeHTML, renderFrames, etc.), descriptions are inconsistent in depth and actionability. Most tools lack comprehensive parameter descriptions, many parameters have minimal or no guidance on constraints, formats, or valid ranges. Input schemas are present but incomplete: several tools accept 'options' objects with nested properties that are not fully documented. Output schemas are largely undocumented. Error handling guidance is absent. The server targets a narrow domain (video generation + TTS) rather than general-purpose agent use, which limits applicability but does not excuse schema gaps. Per-tool scores average 38/100.
Analyze HTML + TTS timing. Outputs JSON with animation info, TTS audio durations, overlap detection, and suggested timing.
Analyze TTS timing without generating audio files.
Assemble per-scene HTML files into one playable document using a timeline JSON. Validates timestamps and detects scene-local/chapter-global mismatches.
Create self-contained HTML with all resources as base64 data URIs.
Clean up temporary TTS audio files.
Create FFmpeg encoder for video output with optional audio mixing.
Generate TTS audio before HTML exists, return exact timestamps. Script is JSON array of {text, pause_after}.
Generate TTS audio with timeline synchronization, returning exact timestamps for each cue.
Nested 'options' objects lack detailed property documentation. Tools like renderFrames, generateTTSWithTimeline, and generateFromScript accept 'options' with nested properties (width, height, fps, engine, voice, model, quality, speed) but provide no descriptions for individual properties, no type constraints, no valid ranges, and no defaults.
Output schemas are undocumented. No tool explicitly declares what fields it returns, their types, or structure. LLMs cannot plan downstream tool calls or extract required data without knowing the response shape.
Error handling and recovery guidance absent. No tool description explains what errors can occur, how to interpret them, or what the LLM should do next (retry, ask user, try alternative tool).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 38 | 2026-07-28+ | v2 |
Scaffold ready-to-fill HTML template (presentation, narrated).
List available HTML templates.
Extract video configuration metadata from HTML meta tags.
Render a single frame at the given time (outputs PNG).
Render all frames for video output at specified FPS and duration.
Export scenes as static PDF pages at specified timestamps.
Interactive HTML preview with play controls, annotation tools, and TTS subtitles.
Render N frames composited into one grid image for visual self-check.
Pre-render lint: font sizes, contrast, sync timestamps, animation quality (appear count, highlight count, fade-out matching, timestamp spread, overlay count).
Parameter constraints missing. Numeric parameters (width, height, fps, duration, timeMs) lack min/max bounds. String parameters (templateName, codec, quality) lack enum constraints or valid value lists. LLMs will pass arbitrary values.
Callback parameters not supported in MCP. renderFrames declares 'onFrame' as a function parameter, which cannot be serialized or invoked over MCP protocol. This tool definition is not implementable as an MCP tool.