MCP-Universe RL: A Framework for Training MCP Tool-Use Agents via Reinforcement Learning

2026-08-22

AI Summary

该研究提出MCP-Universe RL框架，通过强化学习提升大型语言模型工具使用能力，解决了环境隔离和训练调度问题。

Abstract

Reinforcement learning (RL) has become an effective way to improve the tool-use ability of large language models (LLMs), but most existing RL frameworks stop at the policy update. For every new domain, the user is left with two hard systems problems: standing up an isolated environment for each of hundreds of concurrent trajectories and connecting it to training, and scheduling the rollout so that the GPU stays busy across long, multi-turn episodes that spend much of their time stalled on slow tool calls. We present MCP-Universe RL (MCP-U RL), an open-source framework that takes over both. It uses the Model Context Protocol (MCP) as the interface to the environment, so any tool already exposed as an MCP server plugs into training with no RL-specific integration code. It builds the two missing layers once and reuses them across domains: an environment-orchestration layer that provisions, isolates, and recycles the MCP environments over a pluggable container backend, and a rollout-orchestration layer whose staged pipeline overlaps trajectories to keep the GPU busy while episodes wait on tools. A backend-agnostic training layer then applies the update through an existing RL backend, with veRL and slime integrations. With one configuration, changing only the task specification, we train software-engineering, deep-research, and general tool-use agents on gpt-oss-20b and improve task reward in all three.

Problem

• Reinforcement Learning
• Large Language Models
• Tool-Use Ability
• Environment Orchestration
• Rollout Orchestration

Innovation

• Reinforcement Learning
• Model Context Protocol
• Environment Orchestration
• Rollout Orchestration