Jarvis-Brain 0.1.9.15__py3-none-any.whl → 0.1.11.0__py3-none-any.whl

{jarvis_brain-0.1.9.15.dist-info → jarvis_brain-0.1.11.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: Jarvis_Brain
-Version: 0.1.9.15
+Version: 0.1.11.0
 Summary: Jarvis brain mcp
 Requires-Python: >=3.10
 Requires-Dist: beautifulsoup4
@@ -8,3 +8,4 @@ Requires-Dist: curl-cffi
 Requires-Dist: drissionpage
 Requires-Dist: fastmcp
 Requires-Dist: minify-html
+Requires-Dist: pillow

jarvis_brain-0.1.11.0.dist-info/RECORD

@@ -0,0 +1,11 @@
+mcp_tools/__init__.py,sha256=_ZzedxbxmVkdRH00ONOjQR31pZW-B8WPn5P2qE6T-Iw,3012
+mcp_tools/dp_tools.py,sha256=YTl2DEe02UI3qDRfVWxpqVjir6vq9SGw3wY9JFY3DV4,18370
+mcp_tools/main.py,sha256=Fgq2PPuc7XCCwXCSVsyoADB0z1vI_gZ07lnLChcrjjg,1167
+tools/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+tools/browser_manager.py,sha256=zeYcWuzxoohMdnYUoZbRH7axFC_VtV8MsncfN8y0yw0,2023
+tools/browser_proxy.py,sha256=7QdLNSgYV5eYDhlmMiIQ0BkMvXwhPDjTRh4EWgIep6k,9255
+tools/tools.py,sha256=dcxtuv1rKIrqqJud8e7sC8z-hJM8nx1WUUoeQYBIhiY,11098
+jarvis_brain-0.1.11.0.dist-info/METADATA,sha256=D98jNkUXIpAa4IjSGq0bjFcypvADVoBykYxNNAST4gA,264
+jarvis_brain-0.1.11.0.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
+jarvis_brain-0.1.11.0.dist-info/entry_points.txt,sha256=YFQT4xpkUqt5dM5wlKPQQOqcjMuFrT9iuRAzIpAyH7U,51
+jarvis_brain-0.1.11.0.dist-info/RECORD,,

mcp_tools/dp_tools.py

@@ -1,12 +1,17 @@
+"""
+这个文件中提供的工具作为独立的Drissionpage mcp工具
+"""
 import hashlib
 import json
 import os
+import time
 from typing import Any
 
 from fastmcp import FastMCP
 
 from tools.browser_manager import BrowserManager
-from tools.tools import compress_html, requests_html, dp_headless_html, assert_waf_cookie, dp_mcp_message_pack
+from tools.tools import compress_html, requests_html, dp_headless_html, assert_waf_cookie, dp_mcp_message_pack, \
+    compress_html_js, compress_image_bytes
 from tools.browser_proxy import DPProxyClient, DPProxyClientManager
 
 html_source_code_local_save_path = os.path.join(os.getcwd(), "html-source-code")
@@ -94,10 +99,11 @@ def register_get_html(mcp: FastMCP, browser_manager):
         file_name_prefix = hashlib.md5(str(tab.title).encode('utf-8')).hexdigest()
         if not os.path.exists(html_source_code_local_save_path):
             os.makedirs(html_source_code_local_save_path)
-        min_html, compress_rate = compress_html(tab.html)
-        html_str_list = [min_html[i:i + one_turn_max_token] for i in range(0, len(min_html), one_turn_max_token)]
+        # min_html, compress_rate = compress_html(tab.html)
+        min_html = tab.run_js(compress_html_js)
+        # html_str_list = [min_html[i:i + one_turn_max_token] for i in range(0, len(min_html), one_turn_max_token)]
         html_file_list = []
-        for index, html_str in enumerate(html_str_list):
+        for index, html_str in enumerate([min_html]):
             file_name = file_name_prefix + f"_{tab_id}_segment{index}.html"
             abs_path = os.path.join(html_source_code_local_save_path, file_name)
             with open(abs_path, "w", encoding="utf-8") as f:
@@ -299,3 +305,24 @@ def register_scroll_action(mcp: FastMCP, browser_manager):
             browser_port=browser_port,
             tab_id=tab_id,
         )
+
+
+def register_get_screenshot(mcp: FastMCP, browser_manager):
+    @mcp.tool(name="get_tab_screenshot", description="尝试对传入tab页进行截图，并将截图压缩为1M大小png图片，会返回截图保存路径")
+    async def get_tab_screenshot(browser_port: int, tab_id: str) -> dict[str, Any]:
+        _browser = browser_manager.get_browser(browser_port)
+        target_tab = _browser.get_tab(tab_id)
+        if not os.path.exists(html_source_code_local_save_path):
+            os.makedirs(html_source_code_local_save_path)
+        timestamp = int(time.time() * 1000)
+        origin_png = target_tab.get_screenshot(as_bytes="png")
+        compress_png = compress_image_bytes(origin_png)
+        image_path = os.path.join(html_source_code_local_save_path, f"{browser_port}_{tab_id}_{timestamp}.png")
+        with open(image_path, "wb") as f:
+            f.write(compress_png)
+        return dp_mcp_message_pack(
+            message=f"已完成对browser_port={browser_port},tab_id={tab_id}的截屏",
+            browser_port=browser_port,
+            tab_id=tab_id,
+            screenshot_path=image_path
+        )

mcp_tools/main.py

@@ -20,6 +20,7 @@ if "TeamNode-Dp" in enabled_modules:
     register_get_html(mcp, browser_manager)
     register_check_selector(mcp, browser_manager)
     register_pop_first_packet(mcp, browser_manager, client_manager)
+    register_get_screenshot(mcp, browser_manager)
     # 页面交互
     register_click_action(mcp, browser_manager)
     register_scroll_action(mcp, browser_manager)
@@ -29,7 +30,7 @@ if "JarvisNode" in enabled_modules:
 
 
 def main():
-    mcp.run(transport="stdio",show_banner=False)
+    mcp.run(transport="stdio", show_banner=False)
 
 
 if __name__ == '__main__':

tools/browser_proxy.py

@@ -126,6 +126,46 @@ class DrissionPageListenerProxy:
         return attr
 
 
+def check_data_packet(packet: DataPacket, client: DPProxyClient):
+    """
+    封装监听到的数据包，并将其存放在client的packet_queue中
+    :param packet:
+    :param client:
+    :return:
+    """
+    url = packet.url
+    method = packet.request.method
+    data = None
+    if packet.request.hasPostData:
+        data = packet.request.postData
+    domain = urlparse(url).netloc
+    body = packet.response.body
+    body_str = json.dumps(body, ensure_ascii=False, separators=(',', ':'))
+    body_str_list = [body_str[i:i + one_turn_max_token] for i in range(0, len(body_str), one_turn_max_token)]
+    body_completed = True
+    packet_filter = client.packet_filter
+    domain_filter = packet_filter.get("domain_filter", None)
+    method_filter = packet_filter.get("method_filter", ["GET", "POST"])
+    for index, body_str in enumerate(body_str_list):
+        # 如果给了domain_filter并且domain没有在domain_filter中时跳过该数据包
+        if domain_filter and domain not in domain_filter:
+            continue
+        # 如果method没有在method_filter中，则跳过该数据包
+        if method not in method_filter:
+            continue
+        if (index + 1) != len(body_str_list):
+            body_completed = False
+        temp_dict = {
+            "url": url,
+            "body_completed": body_completed,
+            "method": method,
+            "request_data": data,
+            "request_headers": dict(packet.request.headers),
+            "response_headers": dict(packet.response.headers),
+            "response_body_segment": body_str.replace("\\", ""),
+        }
+        client.packet_queue.append(temp_dict)
+
 def check_data_packet(packet: DataPacket, client: DPProxyClient):
     """
     封装监听到的数据包，并将其存放在client的packet_queue中

tools/tools.py

@@ -6,6 +6,90 @@ from DrissionPage import ChromiumPage, ChromiumOptions
 from bs4 import BeautifulSoup
 from curl_cffi import requests
 from lxml import html, etree
+import base64
+from PIL import Image
+import io
+
+compress_html_js = """
+    function getSimplifiedDOM(node) {
+        // 1. 处理文本节点
+        if (node.nodeType === Node.TEXT_NODE) {
+            const text = node.textContent.trim();
+            // 限制文本长度，避免大段文章消耗 token，保留前100个字符通常足够定位
+            return text ? text.slice(0, 100) + (text.length > 100 ? '...' : '') : null;
+        }
+
+        // 2. 过滤无用标签
+        const ignoreTags = ['SCRIPT', 'STYLE', 'NOSCRIPT', 'IFRAME', 'SVG', 'LINK', 'META'];
+        if (ignoreTags.includes(node.tagName)) return null;
+        if (node.nodeType !== Node.ELEMENT_NODE) return null;
+
+        // 3. 过滤不可见元素
+        const style = window.getComputedStyle(node);
+        if (style.display === 'none' || style.visibility === 'hidden' || style.opacity === '0') return null;
+        // 过滤宽高太小的元素（往往是埋点空像素）
+        const rect = node.getBoundingClientRect();
+        if (rect.width === 0 || rect.height === 0) return null;
+
+        // --- 开始构建标签字符串 ---
+        const tagName = node.tagName.toLowerCase();
+        let tagStr = tagName;
+
+        // A. 基础标识符 (ID 和 Class)
+        if (node.id) tagStr += `#${node.id}`;
+        if (node.className && typeof node.className === 'string') {
+            // 过滤掉 Tailwind 等太长且无语义的 class，保留有意义的业务 class
+            // 这里简单处理，全部保留，让 LLM 自己判断
+            const classes = node.className.trim().split(/\s+/);
+            if (classes.length > 0) tagStr += `.${classes.join('.')}`;
+        }
+
+        // B. 关键属性白名单 (这是你指出问题的核心修复)
+        const props = [];
+
+        // 通用重要属性
+        if (node.getAttribute('role')) props.push(`role="${node.getAttribute('role')}"`);
+        if (node.getAttribute('aria-label')) props.push(`aria-label="${node.getAttribute('aria-label')}"`);
+        if (node.getAttribute('title')) props.push(`title="${node.getAttribute('title')}"`);
+
+        // 特定标签的特定属性
+        if (tagName === 'a') {
+            const href = node.getAttribute('href');
+            // 只保留有意义的链接，忽略 javascript:;
+            if (href && !href.startsWith('javascript')) props.push(`href="${href}"`);
+        } else if (tagName === 'input' || tagName === 'textarea' || tagName === 'select') {
+            if (node.getAttribute('type')) props.push(`type="${node.getAttribute('type')}"`);
+            if (node.getAttribute('name')) props.push(`name="${node.getAttribute('name')}"`);
+            if (node.getAttribute('placeholder')) props.push(`placeholder="${node.getAttribute('placeholder')}"`);
+            if (node.disabled) props.push('disabled');
+            if (node.checked) props.push('checked');
+        } else if (tagName === 'button') {
+            if (node.getAttribute('type')) props.push(`type="${node.getAttribute('type')}"`);
+        } else if (tagName === 'img') {
+            if (node.getAttribute('alt')) props.push(`alt="${node.getAttribute('alt')}"`);
+        }
+
+        if (props.length > 0) {
+            tagStr += ` ${props.join(' ')}`;
+        }
+
+        // 4. 递归子节点
+        const children = Array.from(node.childNodes)
+            .map(getSimplifiedDOM)
+            .filter(n => n !== null);
+
+        // 5. 组装输出
+        // 如果没有子节点，也没有ID/Class，也不是输入框/图片/链接，那这个标签可能只是布局用的 div，可以考虑跳过它直接返回子节点内容
+        // 但为了保持结构完整，我们暂时保留它
+        if (children.length === 0) {
+            // 自闭合标签或空标签
+            return `<${tagStr} />`;
+        }
+        return `<${tagStr}>${children.join('')}</${tagName}>`; // 结束标签只保留 tagName 节省 token
+    }
+
+    return getSimplifiedDOM(document.body);
+"""
 
 
 # 使用requests获取html，用于测试是否使用了瑞数和jsl
@@ -107,6 +191,72 @@ def dp_mcp_message_pack(message: str, **kwargs):
         }]
     }
 
+
+def btyes2Base64Img(target_byte):
+    """
+    把byte转为base64，用于传输图片
+    :param target_byte:
+    :return:
+    """
+    return "data:image/png;base64," + base64.b64encode(target_byte).decode()
+
+
+def compress_image_bytes(input_bytes, target_size_mb=1):
+    """
+    压缩图片字节数据到目标大小
+
+    参数:
+        input_bytes: 输入图片的字节数据
+        target_size_mb: 目标大小(MB)，默认1MB
+
+    返回:
+        压缩后的图片字节数据
+    """
+    target_size = target_size_mb * 1024 * 1024  # 转换为字节
+
+    # 从字节数据打开图片
+    img = Image.open(io.BytesIO(input_bytes))
+
+    # 如果是PNG或其他格式，转换为RGB
+    if img.mode in ('RGBA', 'LA', 'P'):
+        img = img.convert('RGB')
+
+    # 初始质量设置
+    quality = 95
+
+    # 先尝试压缩
+    output_buffer = io.BytesIO()
+    img.save(output_buffer, 'JPEG', quality=quality, optimize=True)
+    output_bytes = output_buffer.getvalue()
+
+    # 如果文件仍然太大，逐步降低质量
+    while len(output_bytes) > target_size and quality > 10:
+        quality -= 5
+        output_buffer = io.BytesIO()
+        img.save(output_buffer, 'JPEG', quality=quality, optimize=True)
+        output_bytes = output_buffer.getvalue()
+
+    # 如果降低质量还不够，尝试缩小尺寸
+    if len(output_bytes) > target_size:
+        width, height = img.size
+
+        while len(output_bytes) > target_size and quality > 10:
+            # 缩小10%
+            width = int(width * 0.9)
+            height = int(height * 0.9)
+            img_resized = img.resize((width, height), Image.Resampling.LANCZOS)
+            output_buffer = io.BytesIO()
+            img_resized.save(output_buffer, 'JPEG', quality=quality, optimize=True)
+            output_bytes = output_buffer.getvalue()
+
+    final_size = len(output_bytes) / (1024 * 1024)
+    # print(f"压缩完成！")
+    # print(f"原始大小: {len(input_bytes) / (1024 * 1024):.2f}MB")
+    # print(f"压缩后大小: {final_size:.2f}MB")
+    # print(f"最终质量: {quality}")
+
+    return output_bytes
+
 # todo: 大致盘一下各种判定的逻辑【以下的所有压缩比之间的差距均取“绝对值”】
 #  1. 如果requests、无头、有头获取到的压缩比之间从差距都在15%以内，则认定该页面是静态页面，此时优先使用requests请求
 #  2. 如果requests的status_code为特定的412，或者521，则判定是瑞数和jsl。[此时还有一个特点：requests的压缩比会与其他两种方式获取到的压缩比差距非常大(一两千的那种)]

jarvis_brain-0.1.9.15.dist-info/RECORD

@@ -1,11 +0,0 @@
-mcp_tools/__init__.py,sha256=_ZzedxbxmVkdRH00ONOjQR31pZW-B8WPn5P2qE6T-Iw,3012
-mcp_tools/dp_tools.py,sha256=ugzHnOf43fJ2LutWlDrYplc4TPn8ZLP9Byb85yENvfM,17059
-mcp_tools/main.py,sha256=k5YOuf2USW7vnlFEBxvLSfxPPHP7nrM1J8lLmVEcSuY,1116
-tools/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-tools/browser_manager.py,sha256=zeYcWuzxoohMdnYUoZbRH7axFC_VtV8MsncfN8y0yw0,2023
-tools/browser_proxy.py,sha256=YDBp1bDEX08ev-Bvbuv0TkWz_nkShe95GS7tfd8GBLc,7606
-tools/tools.py,sha256=zYJCvwy7OUmYWKOhHbItSfttAmJLbC-ixXv9oW-TNTA,5033
-jarvis_brain-0.1.9.15.dist-info/METADATA,sha256=g9Bp_gpYdKAXYMaiWgZGesmaLYpF5KxSJ9ijuwdUn8E,242
-jarvis_brain-0.1.9.15.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
-jarvis_brain-0.1.9.15.dist-info/entry_points.txt,sha256=YFQT4xpkUqt5dM5wlKPQQOqcjMuFrT9iuRAzIpAyH7U,51
-jarvis_brain-0.1.9.15.dist-info/RECORD,,