经 AI Skill Hub 精选评估,friday-studio MCP工具 获评「强烈推荐」。这款AI工具在功能完整性、社区活跃度和易用性方面表现出色,AI 评分 8.2 分,适合有一定技术背景的用户使用。
friday-studio MCP工具 是一款基于 TypeScript 开发的开源工具,专注于 MCP协议、Agent框架、工作空间共享 等核心功能。作为 GitHub 开源项目,它拥有活跃的社区支持和持续的版本迭代,代码完全透明可审计,支持本地部署以保护数据隐私。无论是个人使用还是集成到企业工作流,都能提供稳定可靠的解决方案。
friday-studio MCP工具 是一款基于 TypeScript 开发的开源工具,专注于 MCP协议、Agent框架、工作空间共享 等核心功能。作为 GitHub 开源项目,它拥有活跃的社区支持和持续的版本迭代,代码完全透明可审计,支持本地部署以保护数据隐私。无论是个人使用还是集成到企业工作流,都能提供稳定可靠的解决方案。
# 方式一:npm 全局安装 npm install -g friday-studio # 方式二:npx 直接运行(无需安装) npx friday-studio --help # 方式三:项目依赖安装 npm install friday-studio # 方式四:从源码运行 git clone https://github.com/friday-platform/friday-studio cd friday-studio npm install npm start
# 命令行使用
friday-studio --help
# 基本用法
friday-studio [options] <input>
# Node.js 代码中使用
const friday_studio = require('friday-studio');
const result = await friday_studio.run(options);
console.log(result);
# friday-studio 配置说明 # 查看配置选项 friday-studio --config-example > config.yml # 常见配置项 # output_dir: ./output # log_level: info # workers: 4 # 环境变量(覆盖配置文件) export FRIDAY_STUDIO_CONFIG="/path/to/config.yml"
Friday turns natural-language asks into repeatable AI-powered workflows. You describe what you want in chat — "every morning, triage my inbox, draft replies, file real asks as Linear tickets" — and the result is a workspace.yml you can read, version, share, and run on schedule. Run it once, or run it forever.
Why both a chat and a YAML? Prompts decay. Run the same prompt tomorrow and the output drifts; bump the model and your old contracts shift with it. Friday keeps both halves of the loop: conversation is how you build, configuration is how you ship — agents, MCP tools, signals (HTTP, cron, Slack, email, webhook), and the jobs that wire them together. Read the file. Version it. Hand it to a teammate and it runs the same on their machine.
Two surfaces, one runtime:
- Chat in the playground. Friday picks up your installed skills, MCP servers, memory, and OAuth credentials, and actually drives your tools instead of describing what it would do. - Run autonomously. Capture the pattern in a workspace.yml, bind it to a signal, and the daemon runs it observably and locally on schedule.
It's for developers and operators who want agents in production — reviewing PRs at 3am, draining their inbox before standup, watching competitor pricing, summarizing a Fathom call into a Linear ticket — without wiring queues, secret stores, schedulers, MCP plumbing, and provider clients from scratch every time.
Prefer the packaged version? hellofriday.ai ships the same daemon and playground as a one-click installer for macOS with the launcher, dependencies, and tray UI bundled. The rest of this README is for working in-tree.
The daemon orchestrates several runtimes — Friday agents can be authored in TypeScript or Python, can drive a real browser, and fan out work over a message bus. Install the four below yourself, then scripts/setup-dev-env.sh handles the rest.
You install these (we won't touch your version managers):
| Tool | Min | Why |
|---|---|---|
| [Node.js](https://nodejs.org/) | 24+ | Playground runs under Vite via npx; @atlas/ui builds with svelte-package. Manage via fnm/volta/nvm — auto-installing system Node clobbers project pins. |
| [git](https://git-scm.com/) | any recent | — |
setup-dev-env.sh handles these — installs if missing or below min, otherwise keeps yours:
| Tool | Min | Why |
|---|---|---|
| [Deno](https://deno.com/) | 2.7.0 | Runs the daemon, CLI, link service, and every TS/Svelte package. |
| [Go](https://go.dev/) | 1.26.0 | Builds the Go services in tools/. macOS auto-install via Homebrew; Linux fails with the install URL (no sudo without consent). |
| [uv](https://docs.astral.sh/uv/) | 0.11.0 | Provisions the managed Python that type: "user" agents spawn under. |
[nats-server](https://nats.io/) | 2.12.0 | Internal message bus the daemon spawns for agent ↔ daemon RPC. Pinned to v2.12.8 for fresh installs. |
[agent-browser](https://www.npmjs.com/package/agent-browser) | latest | CLI the bundled web agent shells out to for headless browsing. Pulls a Chromium build (~150MB) on first run. |
Python 3.12 + friday-agent-sdk | pinned | Pre-warmed into uv's cache so the first user-agent spawn isn't a cold start. |
Docker (optional) skips all of the above — docker compose up runs the whole stack in containers.
You don't need a separate database for local development — the daemon persists state through its embedded JetStream bus. Postgres is only needed for production deployments of the link credential service.
git clone https://github.com/friday-platform/friday-studio
cd friday-studio
bash scripts/setup-dev-env.sh # bootstraps deno+go+uv+nats+agent-browser, runs deno install, pre-warms Python
setup-dev-env.sh checks each tool's version against the minimums above and keeps your existing toolchain wherever it satisfies them — auto- install only kicks in for missing or stale tools. It also runs `deno install`, writes the daemon envfile, and pre-warms the uv cache. Re-run it whenever the pinned friday-agent-sdk version bumps in tools/friday-launcher/paths.go.
If the script ends with a ⚠ Skipped: block, follow the per-tool hint and re-run. Common case: agent-browser install fails on a system Node — switch to a user-scoped Node manager (fnm/volta/nvm) and re-run to enable web/browser agents.
set -a . "${FRIDAY_HOME:-$HOME/.friday/local}/.env" 2>/dev/null \ || . "$HOME/.atlas/.env" 2>/dev/null || true set +a curl -X POST ${FRIDAY_TLS_CA:+--cacert "$FRIDAY_TLS_CA"} \ "${FRIDAYD_URL:-http://localhost:8080}/review-pr" \ -d '{"pr_url": "https://github.com/your-org/your-repo/pull/42"}'
The workspace declares an HTTP signal at `/review-pr`, an LLM agent with
the `gh` MCP server, and a system prompt scoped to your team's review
checklist. Wire it to GitHub via `webhook-tunnel` for a public URL.
**Cron — drain my inbox before 9am:**
yaml
Want it all in containers?
docker compose up
Brings up the daemon, playground, link credential service, PTY server, and webhook tunnel. Ports default to 1xxxx to avoid host collisions — see docker-compose.yml.
```bash cp .env.example .env
```
The example file documents every variable the daemon reads — provider keys, proxies, OAuth/GitHub App credentials, integration tokens. The minimum to run a real agent is one LLM provider key.
AI编排框架,架构完整度高。MCP+Agent+Skills三层设计先进,社区活跃度好。工作空间共享特性创新。
该工具使用 NOASSERTION 协议,商用场景请仔细阅读协议条款,必要时咨询法律意见。
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。
📄 NOASSERTION — 请查阅原始协议条款了解具体使用限制。
AI Skill Hub 点评:friday-studio MCP工具 的核心功能完整,质量优秀。对于AI爱好者来说,这是一个值得纳入个人工具库的选择。建议先在非生产环境试用,再逐步推广。
| 原始名称 | friday-studio |
| 原始描述 | 开源MCP工具:Agent Harness Platform — shareable workspaces, MCP tools, skills, memory, and cr。⭐82 · TypeScript |
| Topics | MCP协议Agent框架工作空间共享AI编排技能市场 |
| GitHub | https://github.com/friday-platform/friday-studio |
| License | NOASSERTION |
| 语言 | TypeScript |
收录时间:2026-05-18 · 更新时间:2026-05-19 · License:NOASSERTION · AI Skill Hub 不对第三方内容的准确性作法律背书。