Outcrawl

io.github.outcrawlai由 GitHub 账号 outcrawlai 发布v1.0.1更新于 Nov 3, 2026

面向AI智能体的网络基础设施：抓取、爬取和搜索网络，运行浏览器和智能体。

AI 翻译由 AI 模型根据下方的原文描述翻译，可能有误。

原文描述

Web infrastructure for AI agents: scrape, crawl and search the web, run browsers and agents.

已验证Streamable HTTP可网页运行网页搜索与抓取浏览器自动化AI 与机器学习

概览

Outcrawl 的托管 MCP 服务（api.outcrawl.ai）；需要 OAuth 登录；Streamable HTTP。

暂无 AI 概览：以下信息来自注册中心条目和服务本身。

端点
api.outcrawl.ai
最近检查

安装

在 SourceWeft 中

  1. 打开 控制台中的 Outcrawl，将其添加到工作区。
  2. 连接 所需的 凭据。它们会被私密存储，绝不会在公开页面上展示。
  3. 为需要使用其工具的对话启用该服务。

Web executable，通过 Streamable HTTP。 远程服务在工作区中配置后即可从网页运行时运行。

交给智能体安装

把这段提示词粘贴到 Claude Code、Codex、Cursor 等智能体，或 SourceWeft 对话中。智能体会读取该服务器的安装指南，先向你展示发布者、端点和登录方式，经你同意后再把它添加到你的客户端。密钥不会写进任何命令。

阅读 https://sourceweft.com/mcp/io.github.outcrawlai%2Foutcrawl/install.md，并按其中说明把这个 MCP 服务器接入我的客户端。修改任何配置之前，先告诉我它的发布者、端点和登录方式，等我确认。不要把 API 密钥写进命令或我会提交的文件；问我密钥应从哪里获取。

查看智能体所依据的安装指南

其他 MCP 客户端

把它添加到你客户端的 mcpServers 配置中。 客户端会在首次连接时完成 OAuth 登录。

{
  "mcpServers": {
    "outcrawl": {
      "type": "http",
      "url": "https://api.outcrawl.ai/mcp"
    }
  }
}

README

[Outcrawl]

Outcrawl MCP

Give your AI agent the web. Outcrawl is web infrastructure for AI agents: connect it once, then ask in plain words. Your agent reads any page as clean markdown or JSON, crawls a whole site, searches the web with the results already read, watches pages for changes, and hands a multi-step task (a login, a form, a flow across pages) to an agent working in a real browser.

  • Server: https://api.outcrawl.ai/mcp (streamable HTTP)
  • Sign-in: OAuth in your browser on first use. No key to copy, nothing to paste into a config file.
  • 30 tools: scrape, crawl, search, agent runs, monitors, profiles, recorded sessions, usage, credits and workspace rules. Tool reference

The server is hosted by Outcrawl. There is no local package to install or run. You pick one workspace when you sign in, every call spends that workspace's credits, and every result reports what it cost. The Free plan has 1,000 credits a month; a page load is 1 credit. Pricing

Connect your agent

Claude Code

bash
claude mcp add --transport http outcrawl https://api.outcrawl.ai/mcp

The first tool call opens a browser to sign in and pick a workspace.

Or install the plugin, which adds the server and a skill that teaches Claude which tool to reach for:

text
/plugin marketplace add outcrawlai/outcrawl-mcp/plugin install outcrawl@outcrawl

Cursor

[Add to Cursor]

Or add it in Cursor's MCP settings (~/.cursor/mcp.json):

json
{ "mcpServers": { "outcrawl": { "url": "https://api.outcrawl.ai/mcp" } } }

Codex

bash
codex mcp add outcrawl --url https://api.outcrawl.ai/mcpcodex mcp login outcrawl

Or in ~/.codex/config.toml:

toml
[mcp_servers.outcrawl]url = "https://api.outcrawl.ai/mcp"

Gemini CLI

bash
gemini extensions install https://github.com/outcrawlai/outcrawl-mcp

Any other MCP client

Give the client the URL https://api.outcrawl.ai/mcp and choose OAuth. It registers itself and signs you in through your browser on first use:

json
{ "mcpServers": { "outcrawl": { "url": "https://api.outcrawl.ai/mcp" } } }
A client with no OAuth support

Create an API key in the console at https://outcrawl.ai/console/keys and send it as a header:

json
{  "mcpServers": {    "outcrawl": {      "url": "https://api.outcrawl.ai/mcp",      "headers": { "Authorization": "Bearer ock_YOUR_KEY_ID.YOUR_SECRET" }    }  }}

This is the one place an API key belongs in an MCP config; every client above signs in instead.

Tools

GroupToolsWhat they do
Read the weboutcrawl_scrape, outcrawl_crawl, outcrawl_searchOne known URL to markdown, HTML, links, a screenshot or JSON; a whole site from a seed URL; a web search with every hit read in full
Agentsoutcrawl_agent and 9 moreGive a task to an agent in a real browser under caps you set, follow its steps, read its records as they land, cancel it, hand its browser to a person, pass it a file, list its downloads
Monitorsoutcrawl_monitor_create, outcrawl_monitor_list, outcrawl_monitor_deleteWatch a URL on a schedule and notify Discord, Slack or a webhook when it really changes
Profilesoutcrawl_profile_create, outcrawl_profile_list, outcrawl_profile_get, outcrawl_profile_deleteKeep cookies and site storage so a login survives across runs, optionally pinned to a country, region and city
Sessionsoutcrawl_sessions_list, outcrawl_sessions_get, outcrawl_session_network, outcrawl_session_request, outcrawl_session_downloads, outcrawl_session_deleteEvery session is recorded: its replay URL, network and console digest, any request in full, its downloads
Accountoutcrawl_usage, outcrawl_credits, outcrawl_rules_get, outcrawl_rules_setUsage over a date range, the credit pool, and workspace rules written in plain sentences

Pick the cheapest tool that answers: a scrape to read a page you know, a search to find pages, a crawl to cover a site, and an agent only when the answer is behind a login, a form or a multi-step flow. An agent run costs far more than a scrape for the same text.

Try it

Read https://outcrawl.ai/pricing and list the plans with their prices.

Search for the latest Model Context Protocol changes and summarize the top three.

Crawl the docs at https://outcrawl.ai/docs, at most 20 pages, and list the page titles.

Watch https://example.com/changelog every hour and post to my Slack webhook when it changes.

What is in this repo

FileFor
.claude-plugin/ and .mcp.jsonClaude Code plugin and marketplace
plugin.json and mcp.jsonAgent Plugins package (Cursor, ChatGPT, Codex)
.cursor-plugin/plugin.jsonCursor Marketplace
gemini-extension.jsonGemini CLI extension
server.jsonThe official MCP registry entry, io.github.outcrawlai/outcrawl
skills/outcrawl/SKILL.mdThe skill every package ships: which tool answers which question, and how to keep spend down

This repo only packages the connection; the server is hosted by Outcrawl.

Links

MIT licensed.

来源：README.md，提交 83a7040

工具

0
该服务器需要登录，因此 SourceWeft 无法在没有账号的情况下读取其工具，也未收录。已连接账号所能看到的内容可能因账号及其权限而异。

版本历史

1
  1. v1.0.1最新Nov 3, 2026

评分

已满足适用检查项的 85%

该评分衡量此收录条目的完整程度，依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价，也不表示服务器安全或当前可用：可连通性是带日期的检查结果，评分也不包含安装量或用户评价。

  • 近 30 天内可连通必需25 分
  • 已声明认证方式必需10 分
  • 已列出工具15 分

    请让服务器在无凭据时响应 tools/list，或发布可在沙箱中无凭据启动的软件包。

  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分