Poco

Manus-like autonomy, powered by cloud-based Claude Code

An intelligent agent harnessing cloud-based Claude Code to realize a Manus-like autonomous experience.

English | 中文

---

Positioning

Poco turns Claude Code into a cloud-native, autonomous coworker. Built on cloud-based Claude Code, it delivers a Manus-like autonomous experience while keeping humans in the loop, and it plugs into the MCP/Skills ecosystem for rich extensibility.

Why Poco

• Full Claude Code experience: everything Claude Code can do, not just coding
• General-purpose agents via Skills/MCP: connect tools and data to handle docs, organization, and analysis
• Cloud execution with human control: queue tasks in parallel, confirm before key actions

What you can do

• Give a goal, get an executable plan with live progress
• Let agents read, write, and organize real files safely
• Run multiple tasks in parallel across desktop and mobile

Ecosystem

Poco embraces the MCP/Skills ecosystem so you can connect tools, data sources, and workflows as building blocks. Bring your own MCP servers or Skills to extend what agents can do without reinventing the platform.

Product Gallery

Demo1: use three prompts to build a classic google game with doubao's head 😂

Demo2: AskUserTool and file operations

Demo3: poco can handle different types of files.

Demo4: works fine on mobile devices.

Community

Join our WeChat group for updates and discussion:

---

Quickstart (Docker Compose)

Recommended for first-time setup (prepares .env, directories, permissions, pulls executor image, and creates the bucket; uses local rustfs by default):

./scripts/quickstart.sh

After running, edit .env and set required values (e.g. ANTHROPIC_AUTH_TOKEN), then start services if you used --no-start.

Manual start:

docker compose up -d

If you want a lighter setup and use Cloudflare R2 (or any external S3-compatible storage), configure S3_ENDPOINT / S3_ACCESS_KEY / S3_SECRET_KEY / S3_BUCKET in .env (bucket must already exist), then run:

docker compose -f docker-compose.r2.yml up -d

Note: The Executor Manager dynamically creates executor containers. The bootstrap script pulls the executor image by default. If you skip the script, you can pull it manually:

docker pull ghcr.io/poco-ai/poco-executor:latest

Default URLs:

• Frontend: http://localhost:3000
• Backend: http://localhost:8000 (/docs)
• Executor Manager: http://localhost:8001 (/docs)

Docs:

• Docker Compose: docs/en/docker-compose.md
• Environment variables: docs/en/configuration.md