Build an agent tool
in minutes.

2. Mistake check

• ✓ Looks good. No mistakes found.

Why this matters: Writing tool descriptions and schemas that agents use reliably.

3. Export

{
  "type": "function",
  "function": {
    "name": "search_web",
    "description": "Search the public web and return the top results with title, URL and snippet. Use for current events or facts not in your training data. Do not use for URLs you already have; use fetch_url instead.",
    "parameters": {
      "type": "object",
      "properties": {
        "query": {
          "type": "string",
          "description": "The search query, in natural language or keywords. Example: \"MCP server spec 2026\"."
        },
        "max_results": {
          "type": "integer",
          "description": "How many results to return, 1-10. Defaults to 5."
        },
        "recency": {
          "type": "string",
          "description": "Only return results from this period.",
          "enum": [
            "day",
            "week",
            "month",
            "year",
            "any"
          ]
        }
      },
      "required": [
        "query"
      ],
      "additionalProperties": false
    }
  }
}

The zip has a TypeScript and a Python MCP server (stdio) with README, package files and all three definitions. Everything is generated in your browser; nothing is sent to us. Drafts save in this browser automatically.