JSON Schema Preview

Read, validate and refactor JSON Schemas — including private ones — without leaving VS Code.

Install · Documentation · Commands · CLI · Changelog

Generate a schema from a JSON file, open the schema viewer beside it, live-edit the schema's title and watch the docs refresh, configure how they render, then generate TypeScript types from the same schema

_{One continuous take: infer a schema from data, preview it side by side, live-edit it, configure the render, then generate types — all from the editor toolbar.
Per-feature demos are on the documentation site.}

Why this and not the built-in JSON support?

VS Code already validates JSON against a schema. This extension is for everything that starts after that:

	Built-in JSON/YAML support	JSON Schema Preview
Validate JSON against a schema	✅	✅
Validate YAML, TOML and JSONL data	Partly (YAML via a separate extension)	✅
Schemas behind authentication	❌ red squiggle, no IntelliSense	✅ GitHub OAuth, Bearer, Basic
Read a schema as documentation	❌	✅ live preview panel
Is this schema change breaking?	❌	✅ classified diff + CI gate
Refactor a schema (extract, inline, rename)	❌	✅
Generate a schema from data, or types/data from a schema	❌	✅
Test a schema against pinned cases	❌	✅ `*.schema.test.json`
Works headlessly in CI	❌	✅ `jstk` CLI
Answers questions for your AI agent	❌	✅ MCP + language model tools

Features

Read


Preview	Renders a schema as live, navigable HTML documentation in a side panel, scroll-synced with the editor. Ctrl+K V.
Outline	The Outline view, breadcrumbs and Go-to-Symbol show the schema's shape — properties, types, what's required — not the document's.
`$ref` navigation	Ctrl+click a `$ref` to jump to it; hover for a summary.
`$ref` graph	A bird's-eye diagram of which definitions reference which, and every external document pulled in.

Check


Validate	JSON, JSONC, JSONL, YAML and TOML against a bound schema, with errors in the Problems panel and one-click quick fixes.
Lint	Schema-quality findings, including examples and defaults that contradict their own subschema — the bug no language server looks for.
Schema tests	Pin the documents a schema must accept and must reject in a `*.schema.test.json` file. Runs in the editor, in the workspace sweep, and in CI.
Workspace sweep	One command answers "is my repo green?" across data files, schema quality and schema contracts.

Evolve


Diff & compatibility verdict	Classifies every change as breaking or not, against Git `HEAD`, a file or a URL — with a CodeLens showing the count while you edit.
CI gate	The same verdict headlessly: `npx json-schema-toolkit diff old.json new.json --check --strict`.
Refactor	Extract a subschema to `$defs`, inline a `$ref`, rename a definition across every reference, find all references, and delete definitions nothing uses.
Draft migration	Convert between draft-07, 2019-09 and 2020-12.

Generate


Schema from data	Infer a starting schema from a document you already have.
Data from schema	Valid sample instances — one, or many as a JSON array or JSONL. Every instance is validated before you see it.
Types from schema	TypeScript, Python, Go, Rust, Java, C#, Kotlin, Swift, Dart and C++.
Bundle / dereference	Flatten a multi-file schema into one self-contained document.

Reach


Private schemas	Authenticate against GitHub, Artifactory or any HTTPS endpoint — then cache the schema locally so VS Code's own language servers see it too. No more red squiggle.
Catalogs	Bind from SchemaStore or your own private catalog.
Offline	Falls back to the last cached copy when the network is unreachable, and says so.

📖 Full feature list and command reference →

Getting started

Install from the Marketplace — or ext install samdidos.json-schema-preview.
Open a JSON Schema and press Ctrl+K V to preview it.
Open a data file, run JSON Schema: Bind Schema…, then Ctrl+K J to validate it.

A guided walkthrough opens on first install. Python is optional — the richer renderer uses json-schema-for-humans when it's available, and a built-in dependency-free renderer otherwise.

Getting Started guide →

Private & authenticated schemas

When a schema sits behind authentication, VS Code's language server can't fetch it — you get a red squiggle and IntelliSense goes dark. This extension authenticates on your behalf (GitHub OAuth, Bearer token, or Basic auth) and can cache the schema locally so the built-in language servers read it too.

Credentials go to your OS keychain via VS Code's Secret Storage, and are only ever sent to the host they were saved for.

Authentication guide →

Command-line interface

The same core ships as a standalone CLI, json-schema-toolkit (command: jstk) — validate, lint, diff with a compatibility gate, bundle, migrate, infer, sample, and run schema tests, with no VS Code involved.

npx json-schema-toolkit diff api.v1.json api.v2.json --check --strict
npx json-schema-toolkit test contracts/*.schema.test.json

CLI guide →

For AI agents

Agents get schema questions wrong in exactly the places this project already answers them correctly — whether a change is breaking, whether a document validates. So the deterministic engines are exposed as tools, over two surfaces generated from one definition:

In VS Code — registered as language model tools, so any agent in the editor can call validate, lint, diff, bundle, infer, sample, coverage and test instead of guessing.

Anywhere else — jstk mcp serves the same tools over MCP, so Claude Code, other IDEs, or a CI bot can use them too:

// .mcp.json
{ "mcpServers": { "json-schema": { "command": "npx", "args": ["json-schema-toolkit", "mcp"] } } }

Optional AI-assisted authoring works the other way round: draft the descriptions your schema is missing, explain a validation error in plain language, draft a schema from a sentence, or generate realistic sample data.

It is off by default, and it never asks you for an API key: model access goes through VS Code's own Language Model API, so your configured provider does the work. Everything a model produces is put through this project's own engines — it must parse, compile under Ajv, lint clean, and be able to produce a valid instance — before it is offered, and nothing is written to a file without a preview.

AI assistance guide →

Configuration

All settings live under jsonschema.* (User, Workspace or Folder scope). Customise the renderer with a .json-schema-preview-config.json in your workspace root or the jsonschema.config setting — the file wins when both are present.

Full settings reference →

Security, privacy & accessibility

Zero telemetry — nothing about your usage is collected or transmitted, ever. There is no opt-in, because the capability isn't in the code.

Hardened webviews — nonce-based CSP and HTML-escaped schema content in every panel.

Offline-friendly — falls back to the last cached schema when the network is unreachable, and tells you it did.

Accessible by default — every injected control is keyboard-operable and screen-reader-labelled; state is never colour-only.

Security & privacy guide → · Report a vulnerability

Contributing

This project is spec-driven: every change traces to a requirement in specs/, and npm run verify is the single gate that runs locally and in CI. See CONTRIBUTING.md and AGENTS.md — the latter is the guide for AI coding agents working on this repo.

Engineering quality gates