mesh-ops
Cross-adapter polygon modeling recipe pack for the shared cross-DCC modeling vocabulary. Structured RECIPES.yaml entries for primitive creation, hard-surface edits, topology cleanup, UV generation, radial arraying, and material assignment. Reusable by every DCC adapter's mesh-ops
Install
npx skills add https://github.com/dcc-mcp/dcc-mcp-core/tree/main/skills/mesh-ops
claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install dcc-mcp-dcc-mcp-core@llmmart
git clone https://github.com/dcc-mcp/dcc-mcp-core.git
The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole dcc-mcp/dcc-mcp-core collection as a plugin from our marketplace. Git is the plain clone.
Skill manifest
mesh-ops
Cross-adapter polygon modeling recipes for the shared cross-DCC modeling
vocabulary. Every recipe is a DCC-agnostic plan: a validated inputs_schema,
an ordered list of steps that dispatch the canonical mesh_ops__<verb> tool,
a Draft 2020-12 output_contract, and a toolset_profiles: [mesh-ops]
binding.
How to use
recipes__listwithskill: mesh-opsto see the available recipes.recipes__geta recipe to read itsinputs_schema,steps, andoutput_contract.recipes__validatecandidate inputs against the recipe schema before dispatching.recipes__applyto obtain the execution plan, then materialize and dispatch the returnedsteps(see "Dispatch contract" below).
Cross-adapter contract
dcc: any— a recipe is not tied to one host; the same plan applies in every DCC adapter that implements the shared modeling verb.toolset_profiles: [mesh-ops]— the recipe belongs to the mesh-ops toolset carried by each adapter's mesh-ops skill.steps[].tooluses the canonicalmesh_ops__<verb>name, never an adapter-specific slug.inputs_schemais a Draft 2020-12 schema validated by Core without a runtimejsonschemadependency.
Dispatch contract
recipes__apply returns the plan verbatim. Core does three things only:
validate inputs against inputs_schema, attach the output_contract, and
return the steps. It does not substitute ${param} placeholders and does
not apply inputs_schema defaults. The caller must materialize each step:
step.arguments[k] == "${name}" -> inputs[name] if present
else inputs_schema.properties[name].default
Every placeholder therefore names an input property that is either listed in
required or carries a default, so a materialized step never still contains
a ${...} string. The test suite asserts this for every recipe.
Resolve mesh_ops__<verb> through tool_routing in RECIPES.yaml:
tool_routing.tools[<verb>][<adapter>]— the concrete tool that adapter publishes (blender_mesh_ops__bevel_edges,maya_mesh_ops__mirror_mesh,houdini_materials__assign_material, …).null— the adapter publishes no equivalent. Usetool_routing.scripting_fallback[<adapter>]and project the script result onto theoutput_contractyourself.
tool_routing resolves the tool, not the argument shape. Step arguments use
the canonical vocabulary above; projecting them onto an adapter's own
signature is that adapter's job — Maya's combine_meshes takes objects and
name where the canonical recipe says object_names and new_name, and
Maya's loft_sections takes name where the recipe says output_name. When
an adapter cannot express a canonical argument, use the scripting fallback
rather than dropping it.
Output contract
output_contract is a Draft 2020-12 schema, not a label. After the tool runs,
project the observed result onto the declared receipt and validate it. The
contracts are deliberately fail-able:
- every mutating recipe requires
changed: true, so a tool that reports success without touching the mesh cannot pass; get_poly_countrequireschanged: false, so a mutating implementation cannot pass a read-only recipe;- component edits require the affected component list plus a post-operation element count;
auto_uvrequiresuv_count >= 1and a SHA-256uv_digest, so a mesh that ends up without UVs fails the contract — the failure mode that shipped an FBX with 31% of meshes un-UV'd.
Validate with jsonschema.Draft202012Validator, or with Core's dependency-free
validator (dcc_mcp_core._runtime.recipe_schema._RecipeSchemaValidator) on
Python 3.7 hosts.
Adapter-local duplicates
Some adapters ship their own recipe pack (for example Maya's
maya-mesh-ops). adapter_recipe_aliases in RECIPES.yaml maps each
cross-adapter verb to the adapter-local recipe name. When a
recipes__search query matches both, prefer the adapter-local entry
(dcc: <adapter>) — its steps and contract are verified against that host's
real tools — and fall back to the cross-adapter entry (dcc: any) when the
adapter publishes no local recipe.
Read RECIPES.yaml for the full recipe list.
Files (dcc-mcp-core)
-
RECIPES.yaml 30.6 KB
Loading… -
SKILL.md 5.6 KB
--- name: mesh-ops description: >- Cross-adapter polygon modeling recipe pack for the shared cross-DCC modeling vocabulary. Structured RECIPES.yaml entries for primitive creation, hard-surface edits, topology cleanup, UV generation, radial arraying, and material assignment. Reusable by every DCC adapter's mesh-ops skill (maya, blender, houdini, 3dsmax, and custom hosts) before falling back to raw scripting. license: MIT-0 compatibility: "dcc-mcp-core 0.20.29+, Python 3.7+" allowed-tools: ["Read"] metadata: dcc-mcp: dcc: any recall-context: app_type: any domain: modeling workflow_stage: modeling task_category: query layer: domain maintainer: dcc-mcp stage: modeling version: "1.0.0" tags: [modeling, mesh, polygon, topology, recipe, cross-dcc] search-hint: >- modeling recipe, mesh recipe, create primitive, bevel edges, extrude faces, inset, boolean, edge loop, merge vertices, triangulate, cleanup mesh, combine meshes, mirror, loft sections, lathe profile, radial array, auto UV, UV projection, assign material, set pivot intent: "Look up a structured, DCC-agnostic modeling recipe before dispatching an adapter's mesh-ops tools." side-effects: modifies: false creates: false deletes: false targets: [] produces: [modeling_recipe] requires: [] recipes: RECIPES.yaml --- # mesh-ops Cross-adapter polygon modeling recipes for the shared cross-DCC modeling vocabulary. Every recipe is a DCC-agnostic plan: a validated `inputs_schema`, an ordered list of `steps` that dispatch the canonical `mesh_ops__<verb>` tool, a Draft 2020-12 `output_contract`, and a `toolset_profiles: [mesh-ops]` binding. ## How to use 1. `recipes__list` with `skill: mesh-ops` to see the available recipes. 2. `recipes__get` a recipe to read its `inputs_schema`, `steps`, and `output_contract`. 3. `recipes__validate` candidate inputs against the recipe schema before dispatching. 4. `recipes__apply` to obtain the execution plan, then materialize and dispatch the returned `steps` (see "Dispatch contract" below). ## Cross-adapter contract - `dcc: any` — a recipe is not tied to one host; the same plan applies in every DCC adapter that implements the shared modeling verb. - `toolset_profiles: [mesh-ops]` — the recipe belongs to the mesh-ops toolset carried by each adapter's mesh-ops skill. - `steps[].tool` uses the canonical `mesh_ops__<verb>` name, never an adapter-specific slug. - `inputs_schema` is a Draft 2020-12 schema validated by Core without a runtime `jsonschema` dependency. ## Dispatch contract `recipes__apply` returns the plan **verbatim**. Core does three things only: validate `inputs` against `inputs_schema`, attach the `output_contract`, and return the `steps`. It does **not** substitute `${param}` placeholders and does **not** apply `inputs_schema` defaults. The caller must materialize each step: ```text step.arguments[k] == "${name}" -> inputs[name] if present else inputs_schema.properties[name].default ``` Every placeholder therefore names an input property that is either listed in `required` or carries a `default`, so a materialized step never still contains a `${...}` string. The test suite asserts this for every recipe. Resolve `mesh_ops__<verb>` through `tool_routing` in `RECIPES.yaml`: - `tool_routing.tools[<verb>][<adapter>]` — the concrete tool that adapter publishes (`blender_mesh_ops__bevel_edges`, `maya_mesh_ops__mirror_mesh`, `houdini_materials__assign_material`, …). - `null` — the adapter publishes no equivalent. Use `tool_routing.scripting_fallback[<adapter>]` and project the script result onto the `output_contract` yourself. `tool_routing` resolves the tool, not the argument shape. Step arguments use the canonical vocabulary above; projecting them onto an adapter's own signature is that adapter's job — Maya's `combine_meshes` takes `objects` and `name` where the canonical recipe says `object_names` and `new_name`, and Maya's `loft_sections` takes `name` where the recipe says `output_name`. When an adapter cannot express a canonical argument, use the scripting fallback rather than dropping it. ## Output contract `output_contract` is a Draft 2020-12 schema, not a label. After the tool runs, project the observed result onto the declared receipt and validate it. The contracts are deliberately fail-able: - every mutating recipe requires `changed: true`, so a tool that reports success without touching the mesh cannot pass; - `get_poly_count` requires `changed: false`, so a mutating implementation cannot pass a read-only recipe; - component edits require the affected component list plus a post-operation element count; - `auto_uv` requires `uv_count >= 1` **and** a SHA-256 `uv_digest`, so a mesh that ends up without UVs fails the contract — the failure mode that shipped an FBX with 31% of meshes un-UV'd. Validate with `jsonschema.Draft202012Validator`, or with Core's dependency-free validator (`dcc_mcp_core._runtime.recipe_schema._RecipeSchemaValidator`) on Python 3.7 hosts. ## Adapter-local duplicates Some adapters ship their own recipe pack (for example Maya's `maya-mesh-ops`). `adapter_recipe_aliases` in `RECIPES.yaml` maps each cross-adapter verb to the adapter-local recipe name. When a `recipes__search` query matches both, prefer the adapter-local entry (`dcc: <adapter>`) — its steps and contract are verified against that host's real tools — and fall back to the cross-adapter entry (`dcc: any`) when the adapter publishes no local recipe. Read `RECIPES.yaml` for the full recipe list.
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.