MCP & Tool Calling

AgentScope 2.0

agentscope-ai/agentscope

A production-ready, easy-to-use agent framework designed to leverage the reasoning and tool-use abilities of increasingly agentic LLMs without constraining them with strict prompts or opinionated orchestrations.

★ 28.6KStars
⑂ 3.3KForks
333Open issues
PythonLanguage
Apache-2.0License
Q@project.QualityScoreEditorial score

Project screenshots

Screenshot of AgentScope 2.0 Screenshot of AgentScope 2.0

Overview

AgentScope 2.0 is a Python-based multi-agent framework that provides essential abstractions for building, running, and scaling LLM agents. It includes an SDK layer for composing agents from building blocks like ReAct loops, toolkits, and context management, alongside a batteries-included agent service featuring a FastAPI backend and pre-built Web UI for multi-tenant, multi-session applications.

Key features

  • ReAct reasoning-acting loop with structured output and realtime interruption
  • Agentic tool management over Python tools, MCP servers, and skills
  • Automatic context compaction and tool-result offload via middleware
  • Unified event bus streaming multimodal content
  • Fine-grained permission control and Human-in-the-Loop (HITL)
  • Agentic memory with switchable backends (ReMe, Mem0)
  • Isolated workspace and sandbox execution (Docker, E2B, OpenSandbox, Daytona, K8s, etc.)
  • Multi-tenancy and multi-session isolation with FastAPI backend
  • Integration with GitHub MCP Registry and ClawHub
  • Background task offloading and scheduled agent wakeup

Requirements, installation and quick start

From PyPI: `uv pip install agentscope`. From source: `git clone -b main https://github.com/agentscope-ai/agentscope.git`, `cd agentscope`, `uv pip install -e .`.

Usage

To start the agent service backend: clone the repository, navigate to `examples/agent_service`, and run `python main.py`. To start the web UI, open another terminal, navigate to `examples/web_ui`, and run `pnpm install` followed by `pnpm dev`. For SDK usage, developers can define an agent with a system prompt, model, and toolkit, then stream replies asynchronously using the `agent.reply_stream` method.

Model compatibility and use cases

Supports LLM, embedding, and TTS across major providers including OpenAI, Anthropic, Gemini, DashScope, DeepSeek, Moonshot, xAI, and Ollama.

License and risk notes

Released under the Apache License 2.0.

Editorial verification 2026-08-09: repository URL, owner, description, license and repository statistics were reviewed. License metadata: Apache-2.0. README was fetched for the channel draft; re-check repository dependencies, releases and model terms before production use.

Release and maintenance

Recent updates include support for Feishu and Discord channels, MCP & Skill Hub integration, Daytona and K8s-based workspace/sandbox, ReMe long-term memory, Agentic Memory, and distributed, multi-tenancy RAG service.

MemPalace

mempalace/mempalace

★ 58KPython

QwenPaw

agentscope-ai/qwenpaw

★ 33.9KPython

Composio

composiohq/composio

★ 29.5KTypeScript