A general-purpose 2D irregular-polygon nesting (bin-packing) tool exposed as MCP tools for agentic assistants to nest parts onto sheet stock interactively with inline preview rendering and structured placement reports
Four well-named tools with clear, detailed descriptions (150 - 250 chars each). All tools have complete input schemas with typed parameters and descriptions. Tool names follow verb_noun pattern (design_nest, preview_nest, get_nest_report, export_nest). Schemas properly declare types (float, list[dict], bool) and include defaults. However, output schemas are not formally documented, responses are inferred from code. Error handling is minimal; no recovery guidance or categorization. No tool annotations (readOnlyHint/destructiveHint) despite clear risk profiles (export_nest is WRITE, others READ_ONLY). Parameter descriptions lack format constraints (e.g., rotation_step_degrees has no min/max bounds stated). Missing pagination/limits for list responses.
Pack a set of parts onto sheet stock (no files written) and return a summary -- sheets used and per-sheet utilization -- so an agent can cheaply try job specs (part counts, sheet size, rotation step) before asking for a full placement report or file export. Reports MCP progress (a placement heartbeat) on a large job if the caller requested it.
Pack a set of parts and write output files to disk (mirrors the `pyfit` CLI): one DXF per sheet actually used, written to "<output_path>_sheet<N>.dxf". preview=True also writes "<output_path>_sheet<N>.png". Returns the list of files written alongside the full placement report. Reports MCP progress (a placement heartbeat) on a large job if the caller requested it.
Pack a set of parts onto sheet stock and return the full placement report as structured data (sheets used, per-sheet utilization, and every part instance's sheet index, position, rotation, and mirror flag) without writing any files. Reports MCP progress (a placement heartbeat) on a large job if the caller requested it.
Render a quick 2D preview of every sheet's layout (boundary plus every placed part's outline, labeled with utilization) and return them inline as images, so a nesting result can be seen in-conversation before committing to any file export. Reports MCP progress (a placement heartbeat) on a large job if the caller requested it.
Output schemas not formally documented. Code shows return types (dict, list) but no structured schema definition visible for LLM planning. design_nest returns {sheets_used, utilization_by_sheet}; preview_nest returns list of images; get_nest_report returns structured placement data; export_nest returns files + report. LLMs cannot reliably extract fields without explicit schema.
No tool annotations (readOnlyHint/destructiveHint/idempotentHint). export_nest is marked WRITE (file creation) but lacks destructiveHint annotation. design_nest, preview_nest, get_nest_report are READ_ONLY but lack readOnlyHint. Agents cannot infer safety/idempotency without explicit hints.
Parameter constraints missing. rotation_step_degrees (float, default 15.0) has no stated min/max bounds. Valid range is unclear, is 0.1 allowed? 360? 1000? sheet_width/sheet_height have no min bounds (negative sheets?). parts list has no size limit. LLMs will pass invalid values without explicit constraints.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 62 | 2026-07-28+ | v2 |
Error handling lacks recovery guidance. Code comment states 'malformed job/part spec or out-of-range rotation_step_degrees raises ValueError, which FastMCP's dispatcher turns into isError=True tool result', but no actionable error messages visible. LLMs receive raw ValueError without guidance on what to fix or which tool to call next.
parts parameter structure under-documented. Description states 'List of job-spec part dicts, each with: {"name", "quantity", "dxf" (path to DXF file with exactly one closed loop) or "polygon" (inline [[x, y], ...] list), "allow_mirror" (optional, default true)}', but this is inline documentation, not a formal schema. LLMs cannot validate part dict structure without explicit type definitions.