MCP backend
Use when: Orthographic template conflict detection, canonical view policy, rigid-vs-cheat-view decision, front/side/top dimension ratios, source-of-truth negotiation before reconstruction.
Before the first Blender call, read ../text-to-blender/references/mcp-backends.md. Prefer official MCP for shared capabilities; preserve the selected backend for the whole task. Tool names in recipes are client/legacy examples: map them through that reference and its agent-host guide.
Multiview Constraint Solver
Use this before another rebuild when references disagree. A renderer cannot make one rigid mesh satisfy mutually incompatible orthographic dimensions unless you allow view-specific cheats.
Workflow
- Normalize every reference view into a common axis contract:
- front: width X, height Z;
- side: depth Y, height Z;
- top: width X, depth Y;
- back: width X, height Z.
- Measure visible body bboxes, not labels/background/aura unless they are part of the model contract.
- Compare shared axes:
- front width vs top width;
- front/back height vs side height;
- side depth vs top depth.
- If ratios differ beyond tolerance, write a conflict report and select a policy:
front_side_canonical: front brand read and side depth/height are authoritative;front_top_canonical: front brand read and top footprint are authoritative;corrected_templates_required: no rigid rebuild should claim final fit;view_specific_cheats: separate camera/view variants allowed, not one rigid asset.
- Feed the chosen policy into
fit-repair-optimizerbefore any geometry, UV, or lighting work.
Hard gates
- Do not restart modeling until canonical policy exists.
- Do not treat top-view aura/context as body depth.
- Do not average contradictory dimensions unless the user explicitly asks for an approximate stylized compromise.
- Record part-count disagreements separately from bbox disagreements.
Script
scripts/view_constraint_report.pyreads a JSON file of per-view body bboxes and emits shared-axis compatibility and a suggested canonical policy.