AI AgentsInfo
Submit your tool

Checker

Free agent prompt and tool checker

Paste a system prompt and, if you have them, the tool definitions. The server checks the JSON, counts the tokens, and prices one turn and 30 days. It uses the same official model prices as the cost calculator. If you already know the token counts, that calculator is the shorter tool.

Schema notes read . Prices read .

What the server does

  • It checks the tool JSON against the published OpenAI, Anthropic, and MCP shapes, including names, required fields, and the argument schema.
  • It counts the pasted prompt and tool text with tiktoken. When that library has the model, the count is that encoding's count of the pasted text. Other models are labeled as estimates. Message framing is not included.
  • It adds the tool text to the input tokens, then multiplies by the price table. For Claude models with a published tool-use system prompt, that count is added too.
  • It does not write the prompt or the tools to a file, a database, or a log. There is no account.

An array of tools, one tool object, or an object with a tools array. OpenAI function tools use parameters. Anthropic tools use input_schema. MCP tools use inputSchema.

Checking sends the text to this site so the server can count tokens and read the JSON. The text is not saved. Output tokens start at 800 and turns at 10 so the price is visible. Those are not a measured workload.

Sources

Related: agent cost calculator · what is an agent harness · Claude Code and Codex pricing · November harness benchmark · all reports

Cite or embed this

The citation names this page and the date above. The embed is a blockquote with a link back here.

Citation

AI Agents Info. "Free agent prompt and tool checker." Schema notes read November 6, 2026. Prices read November 5, 2026. https://www.aiagentsinfo.com/agent-prompt-checker.

Embed

<blockquote cite="https://www.aiagentsinfo.com/agent-prompt-checker">
<p>Paste a system prompt and optional tool JSON. The server checks the schema, counts tokens, and estimates a 30-day bill from official model prices. The prompt is not stored. Schema notes read November 6, 2026. Prices read November 5, 2026.</p>
<footer><a href="https://www.aiagentsinfo.com/agent-prompt-checker">Free agent prompt and tool checker</a>, AI Agents Info.</footer>
</blockquote>

FAQ

Is the prompt stored?

No. The server reads it to count tokens and check the JSON, then drops it. Nothing is written to a database, a file, or a log. Do not paste secrets you would not put in a request you are debugging.

Is the token count exact?

For gpt-5, gpt-5-mini, and gpt-5-nano, tiktoken has an encoding and the result says so. Newer OpenAI ids in the price list, and every Claude, Gemini, and Grok id, are estimates made with tiktoken's o200k_base encoding. Those providers do not bill with that encoding. Output tokens are the number you type, because this page does not have the reply.

Why is there no tool-call fee?

The price table does not publish a single per-call fee for these models, so none is added. Anthropic does publish a tool-use system prompt length for several Claude models, and that length is added when the row exists. Computer use, code execution, and search grounding stay out when the copied pages do not give a rate.

How is this different from the cost calculator?

The cost calculator stays in the browser. You type token counts, and it multiplies them by the official prices. It does not read a prompt or validate tool JSON. This page sends the text to the server and does both.

Which tool JSON does it accept?

OpenAI function tools, either the Responses shape or the older Chat Completions shape. Anthropic user-defined tools, with name, description, and input_schema. MCP tools, with name and inputSchema. Built-in tools such as web search are reported and left unchecked.