Tool categories

A category-level map of preset AI MCP tool classes. For per-tool details, see the MCP Tools reference.

Copy link to sectionWhy categories matter

The MCP server exposes 40+ tools. Knowing which category to reach for at each stage of a session is the difference between an AI that helps and an AI that thrashes. The default sequence — discovery, constraints, generation, validation, apply — maps onto these categories.

Copy link to sectionFive categories

1. Discovery and retrieval

Read-first context gathering. Call these before generating anything.

Examples: component and preset lookup, semantic search across artifacts, token retrieval.

2. Constraint and guardrail lookup

Call before generation to avoid invalid approaches.

Examples: forbidden primitives, active constraints, intent routing guidance.

3. Generation support

Use when producing code aligned to system artifacts.

Examples: preset-guided code scaffolding, pattern-based generation paths.

4. Validation and quality

Use after generation and before apply.

Examples: drift checks, compliance checks, accessibility and interaction validation.

5. Workflow and governance

Lifecycle-aware write operations.

Examples: proposal-oriented operations, export and context packaging, readiness checks for governed apply.

Copy link to sectionHow to choose tools

The default sequence:

1. Discovery (read)
2. Constraints (read)
3. Generation (write candidate output)
4. Validation (read findings)
5. Proposal or apply (governed write)

Skip stages only when you have a specific reason. Skipping validation in particular tends to surface as drift later.

Copy link to sectionMode-aware behavior

Tool usage adapts to the active compliance mode:

• Observe / Assist — broader flexibility, stronger guidance
• Guard / Enforce — stricter gating, explicit approvals

See adaptive modes for how this is detected and overridden.