// free tool
Will an LLM call your tool correctly?
Paste a tool definition. Get it scored on the things that actually decide whether a model picks it, fills it in properly, and treats it with the right caution.
Runs entirely in your browser — nothing is sent to a server. Accepts OpenAI, Anthropic, MCP and raw JSON Schema tool shapes.
Questions
Which formats does it accept?
OpenAI function tools (both the nested {type:'function'} shape and the flat one), Anthropic tools with input_schema, MCP tools with inputSchema, and raw JSON Schema. It auto-detects which you pasted.
What does the score actually measure?
Whether a model is likely to call your tool correctly. It weights name specificity, description quality, parameter typing and constraints, and whether destructive actions are flagged. Criticals cost 18 points, warnings 7, polish items 2.
Why does it care so much about descriptions?
Because that's what the model reads when deciding between your tools. A tool named getData with the description 'gets data' is functionally invisible in a toolset of twenty — the model can't tell when it applies, so it either skips it or calls it wrongly.
Why flag destructive actions specifically?
Models are measurably more cautious when a description states the consequence. A delete tool whose description doesn't say 'this cannot be undone' is a tool that will eventually be called when it shouldn't have been.
Is my schema sent anywhere?
No. The linter runs entirely in your browser — there's no API call and nothing is stored. You can check with your network tab.
// ready_when_you_are
Let's build your agentic advantage
Tell us what should be automated, connected or sold. We'll map the fastest path from idea to a production system.