AI Skill Hub 强烈推荐:自主AI代理 是一款优质的MCP工具。AI 综合评分 8.0 分,在同类工具中表现稳健。如果你正在寻找可靠的MCP工具解决方案,这是一个值得深入了解的选择。
自主AI代理 是一款遵循 MCP(Model Context Protocol)标准协议的 AI 工具扩展。通过 MCP 协议,它可以让 Claude、Cursor 等主流 AI 客户端直接访问和操作外部工具、数据源和服务,实现 AI 能力的无缝扩展。无论是文件操作、数据库查询还是 API 调用,都可以通过自然语言在 AI 对话中直接触发,极大提升生产效率。
自主AI代理 是一款遵循 MCP(Model Context Protocol)标准协议的 AI 工具扩展。通过 MCP 协议,它可以让 Claude、Cursor 等主流 AI 客户端直接访问和操作外部工具、数据源和服务,实现 AI 能力的无缝扩展。无论是文件操作、数据库查询还是 API 调用,都可以通过自然语言在 AI 对话中直接触发,极大提升生产效率。
# 方式一:通过 Claude Code CLI 一键安装
claude skill install https://github.com/strikersam/autonomous-ai-agency
# 方式二:手动配置 claude_desktop_config.json
{
"mcpServers": {
"--ai--": {
"command": "npx",
"args": ["-y", "autonomous-ai-agency"]
}
}
}
# 配置文件位置
# macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
# Windows: %APPDATA%/Claude/claude_desktop_config.json
# 安装后在 Claude 对话中直接使用 # 示例: 用户: 请帮我用 自主AI代理 执行以下任务... Claude: [自动调用 自主AI代理 MCP 工具处理请求] # 查看可用工具列表 # 在 Claude 中输入:"列出所有可用的 MCP 工具"
// claude_desktop_config.json 配置示例
{
"mcpServers": {
"__ai__": {
"command": "npx",
"args": ["-y", "autonomous-ai-agency"],
"env": {
// "API_KEY": "your-api-key-here"
}
}
}
}
// 保存后重启 Claude Desktop 生效
The wizard helps you choose providers, models, runtimes, a default agent, and a cost policy.
<p align="center"><img src="docs/screenshots/readme/v4-setup-wizard.png" width="92%" alt="Setup Wizard"/></p>
git clone https://github.com/strikersam/autonomous-ai-agency.git
cd autonomous-ai-agency
python -m venv .venv && source .venv/bin/activate
pip install -r backend/requirements.txt
Push to master — CI does the rest:
Required secrets:
RENDER_DEPLOY_HOOK_URL → Render → service → Settings → Deploy Hook
RENDER_BACKEND_URL → your Render service URL
Live demo: - Frontend: https://strikersam.github.io/autonomous-ai-agency/ - API: https://autonomous-ai-agency.onrender.com/docs (pending Render service rename; currently still served at https://local-llm-server.onrender.com/docs until the Render dashboard service is renamed)
Render free tier sleeps after 15 min of inactivity (~30 s cold start). Upgrade to Starter ($7/mo) for always-on in production.
---
Screenshots captured from live deployment. Run pytest tests/e2e/ --screenshot=on to regenerate.
Central settings: defaults, policies, and integrations in one place.
<p align="center"><img src="docs/screenshots/readme/v4-settings.png" width="92%" alt="Settings"/></p>
cp .env.example .env
Minimum viable .env for local development (no MongoDB, no GPU):
STORAGE_BACKEND=sqlite # zero-dependency storage
ADMIN_EMAIL=you@example.com
ADMIN_PASSWORD=changeme
SECRET_KEY=$(openssl rand -hex 32)
NVIDIA_API_KEY=nvapi-... # free at build.nvidia.com — no GPU needed
Full list of variables: docs/configuration.md
// Cursor — settings.json
{
"cursor.ai.openaiBaseUrl": "http://localhost:8000",
"cursor.ai.openaiApiKey": "your-api-key-here"
}
See client-configs/ for Aider, Continue, Zed, VSCode, and Claude Code.
---
| Variable | Default | Description |
|---|---|---|
SECRET_KEY | *(required)* | JWT signing key — openssl rand -hex 32 |
STORAGE_BACKEND | mongo | sqlite for zero-dependency storage |
MONGO_URL | mongodb://localhost:27017 | MongoDB connection string |
OLLAMA_BASE_URL | http://localhost:11434 | Local Ollama server |
LLM_PROVIDER | ollama | ollama · nvidia-nim · deepseek · bedrock · anthropic |
NVIDIA_API_KEY | *(optional)* | Free-tier cloud inference — no GPU required |
AWS_ACCESS_KEY_ID + AWS_SECRET_ACCESS_KEY | *(optional)* | AWS Bedrock |
ANTHROPIC_API_KEY | *(optional)* | Direct Anthropic API |
DEEPSEEK_API_KEY | *(optional)* | DeepSeek cloud API |
GITHUB_TOKEN | *(optional)* | Required for agents that open PRs or read issues |
LANGFUSE_HOST + keys | *(optional)* | Observability traces |
TELEGRAM_BOT_TOKEN | *(optional)* | Remote control via Telegram |
ADMIN_EMAIL + ADMIN_PASSWORD | *(optional)* | First admin — created on first boot |
RUNTIME_DOCKER_ENABLED | false | Enable Docker agent runtime |
curl -X POST http://localhost:8001/api/company/{id}/onboarding/start \ -H "Authorization: Bearer <your-token>" \ -H "Content-Type: application/json" \ -d '{"website_urls": ["https://yourcompany.com"]}' ```
| Workflow | Trigger | What it does |
|---|---|---|
**issue-context-generator.yml** | Any issue opened, or quick-note label added | Fetches the linked URL, generates a grounded prompt + TODO plan, commits docs/context/issue-N.md, opens a **draft PR**, closes the issue. |
**bulk-issue-context.yml** | Manual (workflow_dispatch) | Backfills **all** open issues in one run. Supports dry_run, label exclusions, explicit issue_numbers targeting, and regenerate mode (updates an existing draft PR in place — preserves the PR number). |
**process-quick-note.yml** | Schedule / manual | Picks up a context branch and implements the plan — runs the agentic loop, tests, applies review feedback, and opens its PR as a **draft**. Reuses an existing claude/context-issue-N branch so implementation commits land on the pre-built draft PR. |
We'd rather under-promise. Here's the honest split (see docs/architecture/feature-maturity-matrix.md and docs/support-matrix.md for the full breakdown):
| Area | Maturity | Notes |
|---|---|---|
OpenAI-compatible proxy (/v1/chat/completions) | **Stable** | Used daily by Claude Code, Cursor, Aider, Continue. |
| Multi-provider routing + failover | **Stable** | Bedrock -> NIM -> DeepSeek -> Anthropic -> Ollama chain with per-provider cooldown. |
| Technology scanner (HTML + DNS + TLS + headers) | **Stable** | BuiltWith-parity off-HTML evidence; headless escalation for bot-walled sites is best-effort and needs Chromium installed in the deployment. |
| Task workflow + dispatcher + HITL gates | **Stable** | Persisted state machine, crash-recovery reconciler. |
| Scheduled tasks / persistent schedule store | **Stable** | Durable across redeploys — ScheduleStore works with both Mongo and SQLite; AgentScheduler rehydrates every cadence on boot; APScheduler's worker thread dispatches on_fire coroutines onto the FastAPI main loop via run_coroutine_threadsafe, so 24x7 cadences produce tasks instead of dying on a cross-loop error. |
| Skill bindings + dynamic skill registry | **Stable** | GitHub discovery with ETag caching and rate-limit semaphores. |
| Langfuse observability + local TCO cost model | **Stable** | Cost figures are an estimated commercial-equivalent model, not a billed invoice. |
| Telegram bot remote control | **Beta** | Works; long-running 24x7 deployment paths are still hardening. |
| Multi-agent orchestration (CEO -> specialists) | **Beta** | Plan->execute->verify loop is solid for single-specialist jobs; deep multi-hop swarms are experimental. |
| External agent runtimes (Goose / Hermes / OpenHands) | **Experimental** | Optional sidecars; the Doctor reports them as warnings, never gating, when absent. |
If a screen or capability isn't listed above, treat it as experimental.
---
autonomous-ai-agency 是一个旨在实现自主化 AI 代理协作的项目。它能够模拟企业级的工作流,通过集成多种 AI 模型与工具,构建一个能够自动处理任务、生成文档并进行代码协作的智能代理生态系统。
本项目引入了全新的 Setup Wizard 功能,通过可视化向导帮助用户快速选择 AI Provider、模型、Runtime 以及默认 Agent。同时,系统内置了完善的 Cost Policy(成本策略)管理,确保 AI 代理在执行任务时处于可控的预算范围内。
项目提供了便捷的 Setup Wizard 安装向导。开发者首先需要通过 git clone 获取源码,并使用 python 创建虚拟环境(.venv)进行依赖安装。安装过程中,向导会引导你配置运行环境、选择模型及设置成本策略,确保项目能够快速进入开发状态。
项目的核心配置集中在 Settings 界面,用于统一管理默认设置、策略及各类 Integrations。开发者可以通过配置 .env 文件进行本地开发,支持使用 sqlite 作为零依赖的 STORAGE_BACKEND。此外,系统还支持通过配置 Cursor、Aider、Continue、Zed、VSCode 或 Claude Code 等 AI 编程工具,实现与本地服务的无缝连接。
除了图形化界面,开发者还可以通过 API 直接调用系统功能。例如,可以通过发送 POST 请求至 onboarding 接口,并携带 Authorization Token 与 JSON 数据,实现自动化引导流程,非常适合集成到现有的企业工作流中。
项目内置了高度自动化的 Workflow 模块。例如,`issue-context-generator.yml` 工作流可以监控 GitHub Issue 的状态,当检测到特定标签或事件时,自动抓取相关 URL,生成基于事实的 Prompt 与 TODO 计划,并自动提交文档、创建 Draft PR,实现从问题发现到方案生成的全自动化闭环。
高质量的开源MCP工具,实现AI自动化和智能代理
该工具未明确声明开源协议,商业使用前请联系原作者确认授权范围,避免侵权风险。
AI Skill Hub 为第三方内容聚合平台,本页面信息基于公开数据整理,不对工具功能和质量作任何法律背书。
建议在沙箱或测试环境中充分验证后,再部署至生产环境,并做好必要的安全评估。
总体来看,自主AI代理 是一款质量优秀的MCP工具,在同类工具中具备一定竞争力。AI Skill Hub 将持续追踪其更新动态,建议收藏备用,结合自身场景选择合适时机引入使用。
| 原始名称 | autonomous-ai-agency |
| Topics | aimcpautonomous |
| GitHub | https://github.com/strikersam/autonomous-ai-agency |
| 语言 | Python |
收录时间:2026-06-24 · 更新时间:2026-06-24 · License:未公布 · AI Skill Hub 不对第三方内容的准确性作法律背书。
选择 Agent 类型,复制安装指令后粘贴到对应客户端