Schemyx docs
AI coding agent control reference
Schemyx MCP indexes repo-proven patterns and paradigms. When an agent prepares a task, Schemyx matches it to bounded, compatible context. Cross-agent skill normalization and enforceable generation guardrails are the beta direction.
Foundations
In development
Theme Forge
Turns visual decisions into portable tokens and rule files developers can give to coding agents as explicit input.
UI patterns
In development
Component Forge
Turns custom component decisions into rules AI can follow when generating product interfaces.
Interface layer
Under construction
API Builder
Planned builder for creating contract-first API context so generated code knows the expected endpoints, schemas, auth, and error states.
Foundations
In development
Theme Forge
Turns visual decisions into portable tokens and rule files developers can give to coding agents as explicit input.
Open Theme ForgeThis builder is under construction. The details below describe the planned direction and expected documentation structure, not a completed product surface.
Inputs
- themeConfig.json
- brand colors
- spacing density
- radius scale
- surface contrast
Outputs
- CSS variables
- Tailwind tokens
- component sizing
- surface states
Guardrails
- Preserve background color values unless the theme config changes them.
- Map token values before applying one-off component styles.
- Keep button, card, badge, and panel treatment consistent across pages.
UI patterns
In development
Component Forge
Turns custom component decisions into rules AI can follow when generating product interfaces.
Open Component ForgeThis builder is under construction. The details below describe the planned direction and expected documentation structure, not a completed product surface.
Inputs
- component inventory
- layout patterns
- variant rules
- interaction states
- composition examples
Outputs
- component contract
- usage rules
- layout recipes
- review checklist
Guardrails
- Prefer existing components before creating new UI primitives.
- Document when to use cards, forms, tables, modals, empty states, and page shells.
- Attach visual review criteria so generated screens can be checked against known patterns.
Interface layer
Under construction
API Builder
Planned builder for creating contract-first API context so generated code knows the expected endpoints, schemas, auth, and error states.
This builder is under construction. The details below describe the planned direction and expected documentation structure, not a completed product surface.
Inputs
- resource model
- endpoint map
- auth mode
- request schema
- response schema
Outputs
- API contract
- typed payloads
- handler rules
- integration notes
Guardrails
- Define schemas before generating implementation details.
- Keep auth, validation, and error behavior explicit.
- Document generated contracts where builders and agents can reuse them.
MCP
Control AI coding agents where they write code
Schemyx MCP exposes repo paradigm indexing, task matching during preparation, compatibility filtering, review ambiguity, and guarded source fallback. Cross-agent skills and enforceable generation guardrails are being brought into the same system.
Request Beta AccessTask preparation workflow
- `schemyx_prepare_task`
- `schemyx_search_codebase`
- `schemyx_get_codebase_recipe`
- `schemyx_get_context_pack`
- `schemyx_read_source_guarded`
MCP modes
Local workspaceIndex repo paradigms, match prepared tasks to bounded context, and perform guarded source fallback inside the configured developer workspace.
Governed data planeConnect a team-scoped Schemyx endpoint for account-backed project context without placing customer database credentials in MCP configuration.
Reversible cutoverAdopt the task-preparation path behind a reversible flag while existing MCP read contracts remain available during the beta transition.
What it serves
indexed repo paradigms, task-matched recipes, project rules, relationships, review warnings, and guarded source fallback.
What stays read-only
During task preparation, the MCP server matches bounded context. It does not yet force skill invocation, block model output, or mutate project code.
Why source fallback exists
Task context stays bounded. When evidence is incomplete or ambiguous, the agent can inspect scoped source instead of guessing.
Theme config
Config fields
`themeConfig.json` should stay the source of truth for shared visual behavior. The app maps these values into CSS variables before components consume them.
Current theme fields
fontFamilySets the app typeface used by the layout and shadcn-style primitives.
radiusControls the shared corner radius for cards, panels, and surface framing.
cardPaddingControls interior spacing for repeated card content.
backgroundColorSets the base page color and should stay stable for the current brand direction.
surfaceColorSets the card and panel base color.
accentColorSets highlight, ring, and subtle motion treatment.
Rule system
Rule structure
A rule should tell a builder what to preserve, what it may change, and how output will be reviewed. Rules should be specific enough to guide generated work.
01
Scope
Name the builder, surface, or contract the rule applies to.
02
Constraint
Describe the behavior, token, boundary, or schema that must be followed.
03
Review
Define what confirms the output stayed inside the rule.
Builder workflow
From input to reusable project controls
Forge output should either solve the current task or become a portable standard for future work. Managed promotion, ownership, and approval are part of the governance direction.
Step 1
Collect inputs
Start with the builder inputs: theme values, component rules, or API contract details.
Step 2
Define the controls
Turn those inputs into an explicit config, policy, or contract before building UI or code.
Step 3
Apply guardrails
Check generated output against the builder guardrails before it becomes part of the platform.
Step 4
Promote reusable rules
When a decision repeats, move it into the shared rule system instead of leaving it inside one page.