Loading ToolBench
Text-to-video animation powered by manimgl and multi-agent LLM pipeline
manim-mcp has clear tool naming with action verbs (generate_, edit_, list_, get_, delete_) and well-structured input schemas. However, the tool definitions visible in the provided source only show schema declarations without complete parameter type specifications for all fields or descriptions for every parameter in the full implementation. Output schemas are not documented, and error handling/recovery guidance is minimal. The server exposes write/destructive operations without confirmation patterns. Tool descriptions are adequate (50-100 chars) but lack dependency hints and prerequisites. Parameters like 'quality' and 'format' are enums described inline but lack formal enum constraints in the schema.
Permanently delete an animation and its files.
Edit an existing animation by describing what to change. Pass the render_id from a previous result and natural-language instructions.
Create a Manim animation from a text description. Returns a video URL, render ID, and generated source code.
Get full details for a render, including a fresh download URL.
List past animations with optional status filter and pagination.
Output schemas are not documented. Tools return 'video URL, render ID, and generated source code' but no structured response schema is visible in the code. LLMs cannot plan downstream calls or extract required fields without knowing the response structure.
Destructive tool (delete_render) lacks confirmation/dry-run pattern. Agents can irreversibly delete animations without a confirmation step, risking accidental data loss.
Quality and format parameters use string type with inline descriptions of enums ('low, medium, high, production, fourk') rather than formal enum constraints in the schema. LLMs may hallucinate invalid values.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 64 | 2026-07-28+ | v2 |
| 2026-03-09 | C | 61 | - | v1 |
Error handling is not visible in the provided source. No recovery guidance, categorization of retryable vs fatal errors, or actionable error messages are documented. Agents lack direction when tools fail.
Tool descriptions are brief (50-80 chars) and lack dependency hints. 'Create a Manim animation from a text description' does not explain prerequisites (is LaTeX required? What image formats are valid?) or when to use edit_animation vs generate_animation.
No pagination metadata visible in list_renders schema (limit and offset parameters exist, but no 'total_count' or 'next_cursor' documented in output). Large result sets risk context window exhaustion.
Tool composition risk: if users must call generate_animation first, then optionally edit_animation, then get_render to fetch the full details, the chain is reasonable. However, no guidance is provided on when to call each tool or what each returns, risking agent confusion.