JSON Schema Compatibility Analyzer
Check JSON Schema structure and reviewed OpenAI, Anthropic, Gemini and MCP compatibility rules locally, with precise diagnostics.
Check schema structure and reviewed target rules. Compatibility is a local assessment, not a provider acceptance test.
32 KiB UTF-8, 200 schema nodes and 12 schema levels. JSON only. References are not resolved, patterns are not executed, and input is never modified.
Ready to analyze.
Enable JavaScript to analyze locally.
Check structure before choosing a target
The analyzer uses the project’s static JSON Schema 2020-12 meta-schema validator, then applies reviewed target rules. It checks schemas rather than validating example instance values. The input is never changed. Findings include schema paths and suggestions that explain semantic consequences.
Seven distinct compatibility views
Generic JSON Schema, strict OpenAI function calling, OpenAI Structured Outputs, ordinary Anthropic tools, Gemini schemas, MCP inputSchema and MCP outputSchema appear separately. OpenAI strict objects require all properties and additionalProperties set to false; Structured Outputs also rejects root anyOf. Ordinary Anthropic tools do not inherit OpenAI strict closure rules. MCP inputSchema requires an object root; its outputSchema may describe another instance type.
Understand unknown results
Compatible means no finding within the reviewed local rules, not universal provider support. Unknown / not verified means the project has not established support for a construct, format or resource boundary. The provider subset is deliberately conservative, while generic and MCP schemas retain the wider 2020-12 vocabulary. Model-specific provider limits need separate review.
Bounded local analysis
Paste up to 32 KiB of JSON. Traversal permits 200 schema nodes and 12 schema levels; provider checks use an eight-level reviewed subset. References and recursive behavior are not resolved, no remote resource is fetched, and patterns are never executed. Defaults and examples are data, not additional schemas. Each target shows at most 60 findings with truncation disclosed. Analysis runs in a terminable worker.