// Decision tree for the 11 hex-core MCP tools. Load when picking between search_components, get_component, get_component_schema, resolve_spec, get_recipe, verify_checklist, or any other hex-core tool call.
[HINT] Download the complete skill directory including SKILL.md and all related files
| name | hex-core-mcp-tools |
| description | Decision tree for the 11 hex-core MCP tools. Load when picking between search_components, get_component, get_component_schema, resolve_spec, get_recipe, verify_checklist, or any other hex-core tool call. |
The server exposes 11 tools. Picking the right one first saves 3–5 unnecessary calls.
| Intent | Call |
|---|---|
| "User gave me a brief / spec / goal — what should I build?" | resolve_spec({ brief }) |
| "I already know the feature shape — is there a recipe for it?" | list_recipes() |
| "I have a recipe slug, give me the full step list" | get_recipe({ slug }) |
| "I need a component by name" | get_component({ name }) |
| "I know the component exists, just need props + hints" | get_component_schema({ name }) |
| "User said 'something like a dropdown'" | search_components({ query }) |
| "I installed these, did I miss any internal deps?" | verify_checklist({ components }) |
| "Spin up a fresh project setup" | scaffold_project({ components, theme }) |
| "Get theme tokens in format X" | get_theme({ name, format }) |
| "What themes exist?" | list_themes() |
| "Apply a CSS override to a component" | customize_component({ name, cssOverrides }) |
resolve_spec(brief)
→ top recipe? yes → get_recipe(slug) → install steps → verify_checklist
→ top recipe? no → top components → get_component(name) for each → verify_checklist
get_component when get_component_schema would do. get_component returns the full file source (often >5KB); get_component_schema returns just props + variants + AI hints (<1KB). Use the schema for reasoning about usage. Use the full component only when you're about to write the file.resolve_spec and going straight to search_components. resolve_spec considers recipes and components ranked together; search_components only returns components. For anything that looks like a multi-component feature, start with resolve_spec.verify_checklist after install. The hex add CLI handles transitive deps, but if an agent installed via MCP get_component + Write, it's easy to forget combobox needs command + popover. verify_checklist catches this in one call.get_theme({ format: "css" }) when the user just wanted the HSL values. Use format: "json" for the flat token map — it's the most token-efficient for reasoning.includeExamples: false to get_component by default. Examples are the fastest way to grok API shape. Keep them on unless you're token-budgeting hard.All 11 tools use z.object().strict() — passing an unrecognized key returns a validation error. Don't attempt to sneak in "hint" fields. If a tool doesn't accept a flag, the flag doesn't exist.
Every tool returns content: [{ type: "text", text: "<stringified JSON>" }]. Parse the JSON before acting on it. Error cases for get_component, get_recipe return a text message like Recipe "foo" not found. — check for this before JSON.parse.
hex-core-recipes-workflowhex-core-theming