Du et al.
RL framework mitigates reasoning-tool trade-off in MCP agents
MCPDisCo uses token-masked reward attribution and gradient surgery to jointly improve reasoning and tool use across 70 servers and 527 tools.
Chinese Tech
Kounianhua Du · Jianxing Liu · Shuai Shao · Ke Li · Xing Sun · Weinan Zhang
Shanghai Jiao Tong University · Tencent
Research Digest··2 min read
The authors introduce MCPDisCo, a reinforcement learning framework for agents operating under the Model Context Protocol (MCP).
Why this paper
From Tencent and Shanghai Jiao Tong University
In one line
MCPDisCo's RL framework for MCP-native agents improves success from 15.79% to 26.32% on LiveMCPBench and reduces chain-of-thought tokens by 5.6%.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§