Method execute

src/tools/web/web-fetch.ts:175–272

(args: z.infer<typeof WebFetchArgs>, ctx: ToolContext)

  argsSchema = WebFetchArgs;

  async execute(args: z.infer<typeof WebFetchArgs>, ctx: ToolContext): Promise<ToolResult> {
    if (isPrivateOrLocal(args.url)) {
      return {
        content: `[WEB_FETCH_BLOCKED] ${args.url} is a private/local address. Use browser_navigate if this is your own dev server, or dev_server_log to read its output.`,
        isError: true,
      };
    }

    const controller = new AbortController();
    const timer = setTimeout(() => controller.abort(), args.timeout_ms ?? 30_000);

    try {
      const response = await proxyFetch(args.url, {
        signal: ctx.signal ? mergeAbortSignals(ctx.signal, controller.signal) : controller.signal,
        redirect: 'follow',
        headers: {
          'User-Agent': args.user_agent ?? DEFAULT_UA,
          'Accept': 'text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8',
          'Accept-Language': 'en-US,en;q=0.9',
        },
      });

      if (!response.ok) {
        return {
          content: `[WEB_FETCH_ERROR] HTTP ${response.status} ${response.statusText} for ${args.url}`,
          isError: true,
        };
      }

      const contentType = response.headers.get('content-type') ?? '';

      // Guard against binary payloads. Reading an image/PDF/font/archive as text
      // dumps garbage bytes into the model's context (and burns tokens). Detect
      // by content-type and bail with a useful message instead of the raw bytes.
      const isBinary = /^(image|audio|video|font)\//i.test(contentType)
        || /(application\/(octet-stream|pdf|zip|gzip|x-tar|x-7z|wasm|x-protobuf)|application\/[^;]*\+(zip|octet))/i.test(contentType);
      if (isBinary) {
        const lenHeader = response.headers.get('content-length');
        return {
          content: `URL: ${response.url}\nStatus: ${response.status}\nContent-Type: ${contentType}\n\n[binary content — not fetched as text]\n` +
            `This is a binary file (${contentType}${lenHeader ? `, ${lenHeader} bytes` : ''}). web_fetch returns text only; ` +
            `it won't read images, PDFs, fonts, or archives. If you need to analyze an image, note its URL and describe what you need — don't re-fetch it.`,
          metadata: { url: response.url, status: response.status, contentType, binary: true },
        };
      }

      const body = await response.text();
      const format = args.format ?? 'text';
      const maxChars = args.max_chars ?? 25_000;

      let output: string;
      if (format === 'html' || !contentType.includes('html')) {
        output = body;
      } else if (format === 'markdown') {
        output = stripBase64Images(htmlToMarkdown(body));   // drop base64 token-bombs; keep http image links
      } else {
        output = htmlToText(body);
      }