From 21f6caacd4d08817a5ce3243d36600665c30daba Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E9=9D=9E=E6=B3=95=E6=93=8D=E4=BD=9C?= Date: Sat, 27 Jul 2024 15:00:06 +0800 Subject: [PATCH] feat: enhance the firecrawl tool (#6705) --- .../provider/builtin/firecrawl/firecrawl.py | 13 +- .../provider/builtin/firecrawl/firecrawl.yaml | 5 +- .../builtin/firecrawl/firecrawl_appx.py | 60 +++-- .../provider/builtin/firecrawl/tools/crawl.py | 62 +++-- .../builtin/firecrawl/tools/crawl.yaml | 229 +++++++++++++++--- .../builtin/firecrawl/tools/crawl_job.py | 20 ++ .../builtin/firecrawl/tools/crawl_job.yaml | 37 +++ .../builtin/firecrawl/tools/scrape.py | 45 ++-- .../builtin/firecrawl/tools/scrape.yaml | 159 +++++++++++- .../builtin/firecrawl/tools/search.py | 34 +-- .../builtin/firecrawl/tools/search.yaml | 52 ++++ 11 files changed, 598 insertions(+), 118 deletions(-) create mode 100644 api/core/tools/provider/builtin/firecrawl/tools/crawl_job.py create mode 100644 api/core/tools/provider/builtin/firecrawl/tools/crawl_job.yaml diff --git a/api/core/tools/provider/builtin/firecrawl/firecrawl.py b/api/core/tools/provider/builtin/firecrawl/firecrawl.py index adcb7ebdd6..24dc35759d 100644 --- a/api/core/tools/provider/builtin/firecrawl/firecrawl.py +++ b/api/core/tools/provider/builtin/firecrawl/firecrawl.py @@ -1,22 +1,19 @@ from core.tools.errors import ToolProviderCredentialValidationError -from core.tools.provider.builtin.firecrawl.tools.crawl import CrawlTool +from core.tools.provider.builtin.firecrawl.tools.scrape import ScrapeTool from core.tools.provider.builtin_tool_provider import BuiltinToolProviderController class FirecrawlProvider(BuiltinToolProviderController): def _validate_credentials(self, credentials: dict) -> None: try: - # Example validation using the Crawl tool - CrawlTool().fork_tool_runtime( + # Example validation using the ScrapeTool, only scraping title for minimize content + ScrapeTool().fork_tool_runtime( runtime={"credentials": credentials} ).invoke( user_id='', tool_parameters={ - "url": "https://example.com", - "includes": '', - "excludes": '', - "limit": 1, - "onlyMainContent": True, + "url": "https://google.com", + "onlyIncludeTags": 'title' } ) except Exception as e: diff --git a/api/core/tools/provider/builtin/firecrawl/firecrawl.yaml b/api/core/tools/provider/builtin/firecrawl/firecrawl.yaml index 613a0e4679..a48b9d9f54 100644 --- a/api/core/tools/provider/builtin/firecrawl/firecrawl.yaml +++ b/api/core/tools/provider/builtin/firecrawl/firecrawl.yaml @@ -31,8 +31,5 @@ credentials_for_provider: label: en_US: Firecrawl server's Base URL zh_Hans: Firecrawl服务器的API URL - pt_BR: Firecrawl server's Base URL placeholder: - en_US: https://www.firecrawl.dev - zh_HansL: https://www.firecrawl.dev - pt_BR: https://www.firecrawl.dev + en_US: https://api.firecrawl.dev diff --git a/api/core/tools/provider/builtin/firecrawl/firecrawl_appx.py b/api/core/tools/provider/builtin/firecrawl/firecrawl_appx.py index bfe3e7999d..3b3f78731b 100644 --- a/api/core/tools/provider/builtin/firecrawl/firecrawl_appx.py +++ b/api/core/tools/provider/builtin/firecrawl/firecrawl_appx.py @@ -1,3 +1,4 @@ +import json import logging import time from collections.abc import Mapping @@ -8,6 +9,7 @@ from requests.exceptions import HTTPError logger = logging.getLogger(__name__) + class FirecrawlApp: def __init__(self, api_key: str | None = None, base_url: str | None = None): self.api_key = api_key @@ -25,14 +27,16 @@ class FirecrawlApp: return headers def _request( - self, - method: str, - url: str, - data: Mapping[str, Any] | None = None, - headers: Mapping[str, str] | None = None, - retries: int = 3, - backoff_factor: float = 0.3, + self, + method: str, + url: str, + data: Mapping[str, Any] | None = None, + headers: Mapping[str, str] | None = None, + retries: int = 3, + backoff_factor: float = 0.3, ) -> Mapping[str, Any] | None: + if not headers: + headers = self._prepare_headers() for i in range(retries): try: response = requests.request(method, url, json=data, headers=headers) @@ -47,47 +51,51 @@ class FirecrawlApp: def scrape_url(self, url: str, **kwargs): endpoint = f'{self.base_url}/v0/scrape' - headers = self._prepare_headers() data = {'url': url, **kwargs} - response = self._request('POST', endpoint, data, headers) logger.debug(f"Sent request to {endpoint=} body={data}") + response = self._request('POST', endpoint, data) if response is None: raise HTTPError("Failed to scrape URL after multiple retries") return response def search(self, query: str, **kwargs): endpoint = f'{self.base_url}/v0/search' - headers = self._prepare_headers() data = {'query': query, **kwargs} - response = self._request('POST', endpoint, data, headers) logger.debug(f"Sent request to {endpoint=} body={data}") + response = self._request('POST', endpoint, data) if response is None: raise HTTPError("Failed to perform search after multiple retries") return response def crawl_url( - self, url: str, wait: bool = False, poll_interval: int = 5, idempotency_key: str | None = None, **kwargs + self, url: str, wait: bool = True, poll_interval: int = 5, idempotency_key: str | None = None, **kwargs ): endpoint = f'{self.base_url}/v0/crawl' headers = self._prepare_headers(idempotency_key) - data = {'url': url, **kwargs['params']} - response = self._request('POST', endpoint, data, headers) + data = {'url': url, **kwargs} logger.debug(f"Sent request to {endpoint=} body={data}") + response = self._request('POST', endpoint, data, headers) if response is None: raise HTTPError("Failed to initiate crawl after multiple retries") job_id: str = response['jobId'] if wait: return self._monitor_job_status(job_id=job_id, poll_interval=poll_interval) - return job_id + return response def check_crawl_status(self, job_id: str): endpoint = f'{self.base_url}/v0/crawl/status/{job_id}' - headers = self._prepare_headers() - response = self._request('GET', endpoint, headers=headers) + response = self._request('GET', endpoint) if response is None: raise HTTPError(f"Failed to check status for job {job_id} after multiple retries") return response + def cancel_crawl_job(self, job_id: str): + endpoint = f'{self.base_url}/v0/crawl/cancel/{job_id}' + response = self._request('DELETE', endpoint) + if response is None: + raise HTTPError(f"Failed to cancel job {job_id} after multiple retries") + return response + def _monitor_job_status(self, job_id: str, poll_interval: int): while True: status = self.check_crawl_status(job_id) @@ -96,3 +104,21 @@ class FirecrawlApp: elif status['status'] == 'failed': raise HTTPError(f'Job {job_id} failed: {status["error"]}') time.sleep(poll_interval) + + +def get_array_params(tool_parameters: dict[str, Any], key): + param = tool_parameters.get(key) + if param: + return param.split(',') + + +def get_json_params(tool_parameters: dict[str, Any], key): + param = tool_parameters.get(key) + if param: + try: + # support both single quotes and double quotes + param = param.replace("'", '"') + param = json.loads(param) + except: + raise ValueError(f"Invalid {key} format.") + return param diff --git a/api/core/tools/provider/builtin/firecrawl/tools/crawl.py b/api/core/tools/provider/builtin/firecrawl/tools/crawl.py index b000c1c6ce..08c40a4064 100644 --- a/api/core/tools/provider/builtin/firecrawl/tools/crawl.py +++ b/api/core/tools/provider/builtin/firecrawl/tools/crawl.py @@ -1,36 +1,48 @@ -import json -from typing import Any, Union +from typing import Any from core.tools.entities.tool_entities import ToolInvokeMessage -from core.tools.provider.builtin.firecrawl.firecrawl_appx import FirecrawlApp +from core.tools.provider.builtin.firecrawl.firecrawl_appx import FirecrawlApp, get_array_params, get_json_params from core.tools.tool.builtin_tool import BuiltinTool class CrawlTool(BuiltinTool): - def _invoke(self, user_id: str, tool_parameters: dict[str, Any]) -> Union[ToolInvokeMessage, list[ToolInvokeMessage]]: - app = FirecrawlApp(api_key=self.runtime.credentials['firecrawl_api_key'], base_url=self.runtime.credentials['base_url']) + def _invoke(self, user_id: str, tool_parameters: dict[str, Any]) -> ToolInvokeMessage: + """ + the crawlerOptions and pageOptions comes from doc here: + https://docs.firecrawl.dev/api-reference/endpoint/crawl + """ + app = FirecrawlApp(api_key=self.runtime.credentials['firecrawl_api_key'], + base_url=self.runtime.credentials['base_url']) + crawlerOptions = {} + pageOptions = {} - options = { - 'crawlerOptions': { - 'excludes': tool_parameters.get('excludes', '').split(',') if tool_parameters.get('excludes') else [], - 'includes': tool_parameters.get('includes', '').split(',') if tool_parameters.get('includes') else [], - 'limit': tool_parameters.get('limit', 5) - }, - 'pageOptions': { - 'onlyMainContent': tool_parameters.get('onlyMainContent', False) - } - } + wait_for_results = tool_parameters.get('wait_for_results', True) + + crawlerOptions['excludes'] = get_array_params(tool_parameters, 'excludes') + crawlerOptions['includes'] = get_array_params(tool_parameters, 'includes') + crawlerOptions['returnOnlyUrls'] = tool_parameters.get('returnOnlyUrls', False) + crawlerOptions['maxDepth'] = tool_parameters.get('maxDepth') + crawlerOptions['mode'] = tool_parameters.get('mode') + crawlerOptions['ignoreSitemap'] = tool_parameters.get('ignoreSitemap', False) + crawlerOptions['limit'] = tool_parameters.get('limit', 5) + crawlerOptions['allowBackwardCrawling'] = tool_parameters.get('allowBackwardCrawling', False) + crawlerOptions['allowExternalContentLinks'] = tool_parameters.get('allowExternalContentLinks', False) + + pageOptions['headers'] = get_json_params(tool_parameters, 'headers') + pageOptions['includeHtml'] = tool_parameters.get('includeHtml', False) + pageOptions['includeRawHtml'] = tool_parameters.get('includeRawHtml', False) + pageOptions['onlyIncludeTags'] = get_array_params(tool_parameters, 'onlyIncludeTags') + pageOptions['removeTags'] = get_array_params(tool_parameters, 'removeTags') + pageOptions['onlyMainContent'] = tool_parameters.get('onlyMainContent', False) + pageOptions['replaceAllPathsWithAbsolutePaths'] = tool_parameters.get('replaceAllPathsWithAbsolutePaths', False) + pageOptions['screenshot'] = tool_parameters.get('screenshot', False) + pageOptions['waitFor'] = tool_parameters.get('waitFor', 0) crawl_result = app.crawl_url( - url=tool_parameters['url'], - params=options, - wait=True + url=tool_parameters['url'], + wait=wait_for_results, + crawlerOptions=crawlerOptions, + pageOptions=pageOptions ) - if not isinstance(crawl_result, str): - crawl_result = json.dumps(crawl_result, ensure_ascii=False, indent=4) - - if not crawl_result: - return self.create_text_message("Crawl request failed.") - - return self.create_text_message(crawl_result) + return self.create_json_message(crawl_result) diff --git a/api/core/tools/provider/builtin/firecrawl/tools/crawl.yaml b/api/core/tools/provider/builtin/firecrawl/tools/crawl.yaml index 3861670140..0c5399f973 100644 --- a/api/core/tools/provider/builtin/firecrawl/tools/crawl.yaml +++ b/api/core/tools/provider/builtin/firecrawl/tools/crawl.yaml @@ -3,76 +3,243 @@ identity: author: Richards Tu label: en_US: Crawl - zh_Hans: 爬取 + zh_Hans: 深度爬取 description: human: - en_US: Extract data from a website by crawling through a URL. - zh_Hans: 通过URL从网站中提取数据。 + en_US: Recursively search through a urls subdomains, and gather the content. + zh_Hans: 递归爬取一个网址的子域名，并收集内容。 llm: This tool initiates a web crawl to extract data from a specified URL. It allows configuring crawler options such as including or excluding URL patterns, generating alt text for images using LLMs (paid plan required), limiting the maximum number of pages to crawl, and returning only the main content of the page. The tool can return either a list of crawled documents or a list of URLs based on the provided options. parameters: - name: url type: string required: true label: - en_US: URL to crawl - zh_Hans: 要爬取的URL + en_US: Start URL + zh_Hans: 起始URL human_description: - en_US: The URL of the website to crawl and extract data from. - zh_Hans: 要爬取并提取数据的网站URL。 + en_US: The base URL to start crawling from. + zh_Hans: 要爬取网站的起始URL。 llm_description: The URL of the website that needs to be crawled. This is a required parameter. form: llm + - name: wait_for_results + type: boolean + default: true + label: + en_US: Wait For Results + zh_Hans: 等待爬取结果 + human_description: + en_US: If you choose not to wait, it will directly return a job ID. You can use this job ID to check the crawling results or cancel the crawling task, which is usually very useful for a large-scale crawling task. + zh_Hans: 如果选择不等待，则会直接返回一个job_id，可以通过job_id查询爬取结果或取消爬取任务，这通常对于一个大型爬取任务来说非常有用。 + form: form +############## Crawl Options ####################### - name: includes type: string required: false label: en_US: URL patterns to include zh_Hans: 要包含的URL模式 + placeholder: + en_US: Use commas to separate multiple tags + zh_Hans: 多个标签时使用半角逗号分隔 human_description: - en_US: Specify URL patterns to include during the crawl. Only pages matching these patterns will be crawled, you can use ',' to separate multiple patterns. - zh_Hans: 指定爬取过程中要包含的URL模式。只有与这些模式匹配的页面才会被爬取。 + en_US: | + Only pages matching these patterns will be crawled. Example: blog/*, about/* + zh_Hans: 只有与这些模式匹配的页面才会被爬取。示例：blog/*, about/* form: form - default: '' - name: excludes type: string - required: false label: en_US: URL patterns to exclude zh_Hans: 要排除的URL模式 + placeholder: + en_US: Use commas to separate multiple tags + zh_Hans: 多个标签时使用半角逗号分隔 human_description: - en_US: Specify URL patterns to exclude during the crawl. Pages matching these patterns will be skipped, you can use ',' to separate multiple patterns. - zh_Hans: 指定爬取过程中要排除的URL模式。匹配这些模式的页面将被跳过。 + en_US: | + Pages matching these patterns will be skipped. Example: blog/*, about/* + zh_Hans: 匹配这些模式的页面将被跳过。示例：blog/*, about/* + form: form + - name: returnOnlyUrls + type: boolean + default: false + label: + en_US: return Only Urls + zh_Hans: 仅返回URL + human_description: + en_US: | + If true, returns only the URLs as a list on the crawl status. Attention: the return response will be a list of URLs inside the data, not a list of documents. + zh_Hans: 只返回爬取到的网页链接，而不是网页内容本身。 + form: form + - name: maxDepth + type: number + label: + en_US: Maximum crawl depth + zh_Hans: 爬取深度 + human_description: + en_US: Maximum depth to crawl relative to the entered URL. A maxDepth of 0 scrapes only the entered URL. A maxDepth of 1 scrapes the entered URL and all pages one level deep. A maxDepth of 2 scrapes the entered URL and all pages up to two levels deep. Higher values follow the same pattern. + zh_Hans: 相对于输入的URL，爬取的最大深度。maxDepth为0时，仅抓取输入的URL。maxDepth为1时，抓取输入的URL以及所有一级深层页面。maxDepth为2时，抓取输入的URL以及所有两级深层页面。更高值遵循相同模式。 + form: form + min: 0 + - name: mode + type: select + required: false + form: form + options: + - value: default + label: + en_US: default + - value: fast + label: + en_US: fast + default: default + label: + en_US: Crawl Mode + zh_Hans: 爬取模式 + human_description: + en_US: The crawling mode to use. Fast mode crawls 4x faster websites without sitemap, but may not be as accurate and shouldn't be used in heavy js-rendered websites. + zh_Hans: 使用fast模式将不会使用其站点地图，比普通模式快4倍，但是可能不够准确，也不适用于大量js渲染的网站。 + - name: ignoreSitemap + type: boolean + default: false + label: + en_US: ignore Sitemap + zh_Hans: 忽略站点地图 + human_description: + en_US: Ignore the website sitemap when crawling. + zh_Hans: 爬取时忽略网站站点地图。 form: form - default: 'blog/*' - name: limit type: number required: false label: - en_US: Maximum number of pages to crawl + en_US: Maximum pages to crawl zh_Hans: 最大爬取页面数 human_description: en_US: Specify the maximum number of pages to crawl. The crawler will stop after reaching this limit. zh_Hans: 指定要爬取的最大页面数。爬虫将在达到此限制后停止。 form: form min: 1 - max: 20 default: 5 + - name: allowBackwardCrawling + type: boolean + default: false + label: + en_US: allow Backward Crawling + zh_Hans: 允许向后爬取 + human_description: + en_US: Enables the crawler to navigate from a specific URL to previously linked pages. For instance, from 'example.com/product/123' back to 'example.com/product' + zh_Hans: 使爬虫能够从特定URL导航到之前链接的页面。例如，从'example.com/product/123'返回到'example.com/product' + form: form + - name: allowExternalContentLinks + type: boolean + default: false + label: + en_US: allow External Content Links + zh_Hans: 允许爬取外链 + human_description: + en_US: Allows the crawler to follow links to external websites. + zh_Hans: + form: form +############## Page Options ####################### + - name: headers + type: string + label: + en_US: headers + zh_Hans: 请求头 + human_description: + en_US: | + Headers to send with the request. Can be used to send cookies, user-agent, etc. Example: {"cookies": "testcookies"} + zh_Hans: | + 随请求发送的头部。可以用来发送cookies、用户代理等。示例：{"cookies": "testcookies"} + placeholder: + en_US: Please enter an object that can be serialized in JSON + zh_Hans: 请输入可以json序列化的对象 + form: form + - name: includeHtml + type: boolean + default: false + label: + en_US: include Html + zh_Hans: 包含HTML + human_description: + en_US: Include the HTML version of the content on page. Will output a html key in the response. + zh_Hans: 返回中包含一个HTML版本的内容，将以html键返回。 + form: form + - name: includeRawHtml + type: boolean + default: false + label: + en_US: include Raw Html + zh_Hans: 包含原始HTML + human_description: + en_US: Include the raw HTML content of the page. Will output a rawHtml key in the response. + zh_Hans: 返回中包含一个原始HTML版本的内容，将以rawHtml键返回。 + form: form + - name: onlyIncludeTags + type: string + label: + en_US: only Include Tags + zh_Hans: 仅抓取这些标签 + placeholder: + en_US: Use commas to separate multiple tags + zh_Hans: 多个标签时使用半角逗号分隔 + human_description: + en_US: | + Only include tags, classes and ids from the page in the final output. Use comma separated values. Example: script, .ad, #footer + zh_Hans: | + 仅在最终输出中包含HTML页面的这些标签，可以通过标签名、类或ID来设定，使用逗号分隔值。示例：script, .ad, #footer + form: form - name: onlyMainContent type: boolean - required: false + default: false label: - en_US: Only return the main content of the page - zh_Hans: 仅返回页面的主要内容 + en_US: only Main Content + zh_Hans: 仅抓取主要内容 human_description: - en_US: If enabled, the crawler will only return the main content of the page, excluding headers, navigation, footers, etc. - zh_Hans: 如果启用，爬虫将仅返回页面的主要内容，不包括标题、导航、页脚等。 + en_US: Only return the main content of the page excluding headers, navs, footers, etc. + zh_Hans: 只返回页面的主要内容，不包括头部、导航栏、尾部等。 + form: form + - name: removeTags + type: string + label: + en_US: remove Tags + zh_Hans: 要移除这些标签 + human_description: + en_US: | + Tags, classes and ids to remove from the page. Use comma separated values. Example: script, .ad, #footer + zh_Hans: | + 要在最终输出中移除HTML页面的这些标签，可以通过标签名、类或ID来设定，使用逗号分隔值。示例：script, .ad, #footer + placeholder: + en_US: Use commas to separate multiple tags + zh_Hans: 多个标签时使用半角逗号分隔 + form: form + - name: replaceAllPathsWithAbsolutePaths + type: boolean + default: false + label: + en_US: All AbsolutePaths + zh_Hans: 使用绝对路径 + human_description: + en_US: Replace all relative paths with absolute paths for images and links. + zh_Hans: 将所有图片和链接的相对路径替换为绝对路径。 + form: form + - name: screenshot + type: boolean + default: false + label: + en_US: screenshot + zh_Hans: 截图 + human_description: + en_US: Include a screenshot of the top of the page that you are scraping. + zh_Hans: 提供正在抓取的页面的顶部的截图。 + form: form + - name: waitFor + type: number + min: 0 + label: + en_US: wait For + zh_Hans: 等待时间 + human_description: + en_US: Wait x amount of milliseconds for the page to load to fetch content. + zh_Hans: 等待x毫秒以使页面加载并获取内容。 form: form - options: - - value: 'true' - label: - en_US: 'Yes' - zh_Hans: 是 - - value: 'false' - label: - en_US: 'No' - zh_Hans: 否 - default: 'false' diff --git a/api/core/tools/provider/builtin/firecrawl/tools/crawl_job.py b/api/core/tools/provider/builtin/firecrawl/tools/crawl_job.py new file mode 100644 index 0000000000..fa6c1f87ee --- /dev/null +++ b/api/core/tools/provider/builtin/firecrawl/tools/crawl_job.py @@ -0,0 +1,20 @@ +from typing import Any + +from core.tools.entities.tool_entities import ToolInvokeMessage +from core.tools.provider.builtin.firecrawl.firecrawl_appx import FirecrawlApp +from core.tools.tool.builtin_tool import BuiltinTool + + +class CrawlJobTool(BuiltinTool): + def _invoke(self, user_id: str, tool_parameters: dict[str, Any]) -> ToolInvokeMessage: + app = FirecrawlApp(api_key=self.runtime.credentials['firecrawl_api_key'], + base_url=self.runtime.credentials['base_url']) + operation = tool_parameters.get('operation', 'get') + if operation == 'get': + result = app.check_crawl_status(job_id=tool_parameters['job_id']) + elif operation == 'cancel': + result = app.cancel_crawl_job(job_id=tool_parameters['job_id']) + else: + raise ValueError(f'Invalid operation: {operation}') + + return self.create_json_message(result) diff --git a/api/core/tools/provider/builtin/firecrawl/tools/crawl_job.yaml b/api/core/tools/provider/builtin/firecrawl/tools/crawl_job.yaml new file mode 100644 index 0000000000..78008e4ad4 --- /dev/null +++ b/api/core/tools/provider/builtin/firecrawl/tools/crawl_job.yaml @@ -0,0 +1,37 @@ +identity: + name: crawl_job + author: hjlarry + label: + en_US: Crawl Job + zh_Hans: 爬取任务处理 +description: + human: + en_US: Retrieve the scraping results based on the job ID, or cancel the scraping task. + zh_Hans: 根据爬取任务ID获取爬取结果，或者取消爬取任务 + llm: Retrieve the scraping results based on the job ID, or cancel the scraping task. +parameters: + - name: job_id + type: string + required: true + label: + en_US: Job ID + human_description: + en_US: Set wait_for_results to false in the Crawl tool can get the job ID. + zh_Hans: 在深度爬取工具中将等待爬取结果设置为否可以获取Job ID。 + llm_description: Set wait_for_results to false in the Crawl tool can get the job ID. + form: llm + - name: operation + type: select + required: true + options: + - value: get + label: + en_US: get crawl status + - value: cancel + label: + en_US: cancel crawl job + label: + en_US: operation + zh_Hans: 操作 + llm_description: choose the operation to perform. `get` is for getting the crawl status, `cancel` is for cancelling the crawl job. + form: llm diff --git a/api/core/tools/provider/builtin/firecrawl/tools/scrape.py b/api/core/tools/provider/builtin/firecrawl/tools/scrape.py index 3a78dce8d0..91412da548 100644 --- a/api/core/tools/provider/builtin/firecrawl/tools/scrape.py +++ b/api/core/tools/provider/builtin/firecrawl/tools/scrape.py @@ -1,26 +1,39 @@ -import json -from typing import Any, Union +from typing import Any from core.tools.entities.tool_entities import ToolInvokeMessage -from core.tools.provider.builtin.firecrawl.firecrawl_appx import FirecrawlApp +from core.tools.provider.builtin.firecrawl.firecrawl_appx import FirecrawlApp, get_array_params, get_json_params from core.tools.tool.builtin_tool import BuiltinTool class ScrapeTool(BuiltinTool): - def _invoke(self, user_id: str, tool_parameters: dict[str, Any]) -> Union[ToolInvokeMessage, list[ToolInvokeMessage]]: - app = FirecrawlApp(api_key=self.runtime.credentials['firecrawl_api_key'], base_url=self.runtime.credentials['base_url']) - crawl_result = app.scrape_url( - url=tool_parameters['url'], - wait=True - ) + def _invoke(self, user_id: str, tool_parameters: dict[str, Any]) -> ToolInvokeMessage: + """ + the pageOptions and extractorOptions comes from doc here: + https://docs.firecrawl.dev/api-reference/endpoint/scrape + """ + app = FirecrawlApp(api_key=self.runtime.credentials['firecrawl_api_key'], + base_url=self.runtime.credentials['base_url']) - if isinstance(crawl_result, dict): - result_message = json.dumps(crawl_result, ensure_ascii=False, indent=4) - else: - result_message = str(crawl_result) + pageOptions = {} + extractorOptions = {} - if not crawl_result: - return self.create_text_message("Scrape request failed.") + pageOptions['headers'] = get_json_params(tool_parameters, 'headers') + pageOptions['includeHtml'] = tool_parameters.get('includeHtml', False) + pageOptions['includeRawHtml'] = tool_parameters.get('includeRawHtml', False) + pageOptions['onlyIncludeTags'] = get_array_params(tool_parameters, 'onlyIncludeTags') + pageOptions['removeTags'] = get_array_params(tool_parameters, 'removeTags') + pageOptions['onlyMainContent'] = tool_parameters.get('onlyMainContent', False) + pageOptions['replaceAllPathsWithAbsolutePaths'] = tool_parameters.get('replaceAllPathsWithAbsolutePaths', False) + pageOptions['screenshot'] = tool_parameters.get('screenshot', False) + pageOptions['waitFor'] = tool_parameters.get('waitFor', 0) - return self.create_text_message(result_message) + extractorOptions['mode'] = tool_parameters.get('mode', '') + extractorOptions['extractionPrompt'] = tool_parameters.get('extractionPrompt', '') + extractorOptions['extractionSchema'] = get_json_params(tool_parameters, 'extractionSchema') + + crawl_result = app.scrape_url(url=tool_parameters['url'], + pageOptions=pageOptions, + extractorOptions=extractorOptions) + + return self.create_json_message(crawl_result) diff --git a/api/core/tools/provider/builtin/firecrawl/tools/scrape.yaml b/api/core/tools/provider/builtin/firecrawl/tools/scrape.yaml index 29aa5991aa..598429de5e 100644 --- a/api/core/tools/provider/builtin/firecrawl/tools/scrape.yaml +++ b/api/core/tools/provider/builtin/firecrawl/tools/scrape.yaml @@ -3,7 +3,7 @@ identity: author: ahasasjeb label: en_US: Scrape - zh_Hans: 抓取 + zh_Hans: 单页面抓取 description: human: en_US: Extract data from a single URL. @@ -21,3 +21,160 @@ parameters: zh_Hans: 要抓取并提取数据的网站URL。 llm_description: The URL of the website that needs to be crawled. This is a required parameter. form: llm +############## Page Options ####################### + - name: headers + type: string + label: + en_US: headers + zh_Hans: 请求头 + human_description: + en_US: | + Headers to send with the request. Can be used to send cookies, user-agent, etc. Example: {"cookies": "testcookies"} + zh_Hans: | + 随请求发送的头部。可以用来发送cookies、用户代理等。示例：{"cookies": "testcookies"} + placeholder: + en_US: Please enter an object that can be serialized in JSON + zh_Hans: 请输入可以json序列化的对象 + form: form + - name: includeHtml + type: boolean + default: false + label: + en_US: include Html + zh_Hans: 包含HTML + human_description: + en_US: Include the HTML version of the content on page. Will output a html key in the response. + zh_Hans: 返回中包含一个HTML版本的内容，将以html键返回。 + form: form + - name: includeRawHtml + type: boolean + default: false + label: + en_US: include Raw Html + zh_Hans: 包含原始HTML + human_description: + en_US: Include the raw HTML content of the page. Will output a rawHtml key in the response. + zh_Hans: 返回中包含一个原始HTML版本的内容，将以rawHtml键返回。 + form: form + - name: onlyIncludeTags + type: string + label: + en_US: only Include Tags + zh_Hans: 仅抓取这些标签 + placeholder: + en_US: Use commas to separate multiple tags + zh_Hans: 多个标签时使用半角逗号分隔 + human_description: + en_US: | + Only include tags, classes and ids from the page in the final output. Use comma separated values. Example: script, .ad, #footer + zh_Hans: | + 仅在最终输出中包含HTML页面的这些标签，可以通过标签名、类或ID来设定，使用逗号分隔值。示例：script, .ad, #footer + form: form + - name: onlyMainContent + type: boolean + default: false + label: + en_US: only Main Content + zh_Hans: 仅抓取主要内容 + human_description: + en_US: Only return the main content of the page excluding headers, navs, footers, etc. + zh_Hans: 只返回页面的主要内容，不包括头部、导航栏、尾部等。 + form: form + - name: removeTags + type: string + label: + en_US: remove Tags + zh_Hans: 要移除这些标签 + human_description: + en_US: | + Tags, classes and ids to remove from the page. Use comma separated values. Example: script, .ad, #footer + zh_Hans: | + 要在最终输出中移除HTML页面的这些标签，可以通过标签名、类或ID来设定，使用逗号分隔值。示例：script, .ad, #footer + placeholder: + en_US: Use commas to separate multiple tags + zh_Hans: 多个标签时使用半角逗号分隔 + form: form + - name: replaceAllPathsWithAbsolutePaths + type: boolean + default: false + label: + en_US: All AbsolutePaths + zh_Hans: 使用绝对路径 + human_description: + en_US: Replace all relative paths with absolute paths for images and links. + zh_Hans: 将所有图片和链接的相对路径替换为绝对路径。 + form: form + - name: screenshot + type: boolean + default: false + label: + en_US: screenshot + zh_Hans: 截图 + human_description: + en_US: Include a screenshot of the top of the page that you are scraping. + zh_Hans: 提供正在抓取的页面的顶部的截图。 + form: form + - name: waitFor + type: number + min: 0 + label: + en_US: wait For + zh_Hans: 等待时间 + human_description: + en_US: Wait x amount of milliseconds for the page to load to fetch content. + zh_Hans: 等待x毫秒以使页面加载并获取内容。 + form: form +############## Extractor Options ####################### + - name: mode + type: select + options: + - value: markdown + label: + en_US: markdown + - value: llm-extraction + label: + en_US: llm-extraction + - value: llm-extraction-from-raw-html + label: + en_US: llm-extraction-from-raw-html + - value: llm-extraction-from-markdown + label: + en_US: llm-extraction-from-markdown + label: + en_US: Extractor Mode + zh_Hans: 提取模式 + human_description: + en_US: | + The extraction mode to use. 'markdown': Returns the scraped markdown content, does not perform LLM extraction. 'llm-extraction': Extracts information from the cleaned and parsed content using LLM. + zh_Hans: 使用的提取模式。“markdown”：返回抓取的markdown内容，不执行LLM提取。“llm-extractioin”：使用LLM按Extractor Schema从内容中提取信息。 + form: form + - name: extractionPrompt + type: string + label: + en_US: Extractor Prompt + zh_Hans: 提取时的提示词 + human_description: + en_US: A prompt describing what information to extract from the page, applicable for LLM extraction modes. + zh_Hans: 当使用LLM提取模式时，用于给LLM描述提取规则。 + form: form + - name: extractionSchema + type: string + label: + en_US: Extractor Schema + zh_Hans: 提取时的结构 + placeholder: + en_US: Please enter an object that can be serialized in JSON + human_description: + en_US: | + The schema for the data to be extracted, required only for LLM extraction modes. Example: { + "type": "object", + "properties": {"company_mission": {"type": "string"}}, + "required": ["company_mission"] + } + zh_Hans: | + 当使用LLM提取模式时，使用该结构去提取，示例：{ + "type": "object", + "properties": {"company_mission": {"type": "string"}}, + "required": ["company_mission"] + } + form: form diff --git a/api/core/tools/provider/builtin/firecrawl/tools/search.py b/api/core/tools/provider/builtin/firecrawl/tools/search.py index 0b118aa5f1..e2b2ac6b4d 100644 --- a/api/core/tools/provider/builtin/firecrawl/tools/search.py +++ b/api/core/tools/provider/builtin/firecrawl/tools/search.py @@ -1,5 +1,4 @@ -import json -from typing import Any, Union +from typing import Any from core.tools.entities.tool_entities import ToolInvokeMessage from core.tools.provider.builtin.firecrawl.firecrawl_appx import FirecrawlApp @@ -7,20 +6,23 @@ from core.tools.tool.builtin_tool import BuiltinTool class SearchTool(BuiltinTool): - def _invoke(self, user_id: str, tool_parameters: dict[str, Any]) -> Union[ToolInvokeMessage, list[ToolInvokeMessage]]: - app = FirecrawlApp(api_key=self.runtime.credentials['firecrawl_api_key'], base_url=self.runtime.credentials['base_url']) - - crawl_result = app.search( + def _invoke(self, user_id: str, tool_parameters: dict[str, Any]) -> ToolInvokeMessage: + """ + the pageOptions and searchOptions comes from doc here: + https://docs.firecrawl.dev/api-reference/endpoint/search + """ + app = FirecrawlApp(api_key=self.runtime.credentials['firecrawl_api_key'], + base_url=self.runtime.credentials['base_url']) + pageOptions = {} + pageOptions['onlyMainContent'] = tool_parameters.get('onlyMainContent', False) + pageOptions['fetchPageContent'] = tool_parameters.get('fetchPageContent', True) + pageOptions['includeHtml'] = tool_parameters.get('includeHtml', False) + pageOptions['includeRawHtml'] = tool_parameters.get('includeRawHtml', False) + searchOptions = {'limit': tool_parameters.get('limit')} + search_result = app.search( query=tool_parameters['keyword'], - wait=True + pageOptions=pageOptions, + searchOptions=searchOptions ) - if isinstance(crawl_result, dict): - result_message = json.dumps(crawl_result, ensure_ascii=False, indent=4) - else: - result_message = str(crawl_result) - - if not crawl_result: - return self.create_text_message("Search request failed.") - - return self.create_text_message(result_message) + return self.create_json_message(search_result) diff --git a/api/core/tools/provider/builtin/firecrawl/tools/search.yaml b/api/core/tools/provider/builtin/firecrawl/tools/search.yaml index b1513c914e..29df0cfaaa 100644 --- a/api/core/tools/provider/builtin/firecrawl/tools/search.yaml +++ b/api/core/tools/provider/builtin/firecrawl/tools/search.yaml @@ -21,3 +21,55 @@ parameters: zh_Hans: 输入关键词即可使用Firecrawl API进行搜索。 llm_description: Efficiently extract keywords from user text. form: llm +############## Page Options ####################### + - name: onlyMainContent + type: boolean + default: false + label: + en_US: only Main Content + zh_Hans: 仅抓取主要内容 + human_description: + en_US: Only return the main content of the page excluding headers, navs, footers, etc. + zh_Hans: 只返回页面的主要内容，不包括头部、导航栏、尾部等。 + form: form + - name: fetchPageContent + type: boolean + default: true + label: + en_US: fetch Page Content + zh_Hans: 抓取页面内容 + human_description: + en_US: Fetch the content of each page. If false, defaults to a basic fast serp API. + zh_Hans: 获取每个页面的内容。如果为否，则使用基本的快速搜索结果页面API。 + form: form + - name: includeHtml + type: boolean + default: false + label: + en_US: include Html + zh_Hans: 包含HTML + human_description: + en_US: Include the HTML version of the content on page. Will output a html key in the response. + zh_Hans: 返回中包含一个HTML版本的内容，将以html键返回。 + form: form + - name: includeRawHtml + type: boolean + default: false + label: + en_US: include Raw Html + zh_Hans: 包含原始HTML + human_description: + en_US: Include the raw HTML content of the page. Will output a rawHtml key in the response. + zh_Hans: 返回中包含一个原始HTML版本的内容，将以rawHtml键返回。 + form: form +############## Search Options ####################### + - name: limit + type: number + min: 0 + label: + en_US: Maximum results + zh_Hans: 最大结果数量 + human_description: + en_US: Maximum number of results. Max is 20 during beta. + zh_Hans: 最大结果数量。在测试阶段，最大为20。 + form: form