Decantr AI MCP

Help AI build like it belongs here.

Agent-neutral UI change control for codebases edited by AI.

Website · Get started · Documentation · Discord

Decantr checks the UI change an AI coding agent actually made against the components, styles, interaction rules, and source boundaries already owned by your project. It reports a small set of source-anchored findings without replacing your stack or generating the change itself.

Check your current UI change

No account, setup, or adoption step is required:

npx @decantr/cli@3.11.4 verify

CHANGED UI          PROJECT AUTHORITY       VERIFY              RESULT

What the agent → What your app → Compare the → Pass, attention, actually edited already owns current change or not proven

Bare verify is read-only. It resolves the current Git change, selects exactly one changed UI app only when that choice is provable, and returns at most three consequential findings with source and repair targets.

From one check to a governed loop

Changed-UI Assurance is the zero-setup entry point. Teams can deepen it into an authority-aware workflow when they need durable task context and CI evidence.

1. Observe project-owned UI authority and state what is unknown.
2. Prepare compact, change-scoped context for any coding agent.
3. Verify the resulting diff against project authority and available evidence.
4. Report typed, reproducible evidence for people, agents, and CI.

npx @decantr/cli@3.11.4 scan

npx @decantr/cli adopt --yes # one-time attachment npx @decantr/cli task /feed "add saved actions" npx @decantr/cli verify --full npx @decantr/cli ci init # one-time CI setup

In a monorepo, install once at the workspace root and select the app consistently:

pnpm add -D -w @decantr/cli

pnpm exec decantr scan --project apps/web pnpm exec decantr adopt --project apps/web --yes pnpm exec decantr task /feed "add saved actions" --project apps/web pnpm exec decantr verify --project apps/web

scan and bare verify are read-only. adopt is the explicit one-time write boundary. Use verify --full for Project Health and its evidence or baseline options.

What Decantr treats as UI authority

Routes are evidence, not the product ontology. Decantr can reason about routes, layouts, reusable components, stories, overlays, flows, design-system packages, runtime states, and exact source files.

It evaluates authority on independent axes:

• selected app and production surface authority;
• topology completeness and implementation taskability;
• component inventory and styling authority;
• available runtime evidence.

The deeper workflow reports ready, limited, blocked, or unsupported. A numeric score never hides an unresolved axis. Source declaration and deployment reachability remain separate facts, and tests, fixtures, stories, generated files, build output, and sibling apps never become production authority.

See UI authority and workflow for the complete model.

Local-first by design

• Source inspection and verification run locally.
• Hosted source upload is retired.
• Browser evidence stays in the project unless you move it.
• Existing repository instructions and MCP servers remain project-owned.
• Production source is the first authority; reviewed local law or style mappings may refine it.
• Official Decantr content is guidance and never silently overrides the application.

Decantr does not replace your router, component library, styling system, tests, design tools, editor, or coding agent.

Current release

CLI and verifier 3.11.4, with MCP server 3.11.5, correct component-reuse import advice while preserving the existing report schemas and eight-tool MCP surface. A feature-specific wrapper is no longer treated as an exact shared export merely because its name ends in the same primitive name.

• Component-reuse patch release note
• Change Assurance contract
• 3.11 qualification evidence
• Package support matrix

Supported package surface

The MCP server preserves exactly eight public tools: decantr_project, decantr_contract, decantr_context, decantr_graph, decantr_registry, decantr_verify, decantr_repair, and decantr_contract_write.

Research boundary

Decantr does not currently claim that it materially improves frontier-model outcomes. That causal claim is governed by a separate, unfinished, predeclared research program. Deterministic product qualification and shipped availability do not substitute for paired model evidence or independent human review.

Read the frontier-model lift research program and 4.0 entry criteria for the complete gates and current limitations.

Development

Requires Node.js >=20.19.0 and pnpm >=9.

pnpm install

pnpm build pnpm test pnpm qualification:3-11:changes pnpm benchmark:3-10:validate

Paid research execution is not implied by product release validation. It requires explicit budget approval, configured provider credentials, frozen treatment artifacts, sealed evaluators, and independent review.

Documentation

• Existing apps
• Change Assurance
• AI assistant setup
• Workflow model
• Command surface
• FAQ
• Release notes
• Research and qualification evidence

Historical programs, audits, benchmarks, and release notes remain available as evidence of what was proposed or shipped at that time. They do not override active references or convert regression evidence into a model-lift result.

License

MIT. Test-corpus repositories retain their own licenses and are not redistributed by Decantr.