mcp-reasoner
MCP Reasoner is a project aimed at enhancing Claude Desktop's problem-solving capabilities through the implementation of Beam Search and Monte Carlo Tree Search. It introduces new experimental algorithms for complex reasoning tasks.
What are the new features in version 2.0.0?
Version 2.0.0 introduces experimental reasoning algorithms with alpha implementations of Policy Simulation Layer and Adaptive Exploration Simulator.
Why were mcts-001-alpha and mcts-001alt-alpha removed?
They were removed because they were similar to the base MCTS method and did not show significant improvement in reasoning processes.
What is the purpose of the Policy Simulation Layer?
The Policy Simulation Layer is intended to integrate policy and search in tandem, which is a common approach in many algorithms.
How can I configure MCP Reasoner with Claude Desktop?
You can configure it by adding the MCP Reasoner command and arguments to the Claude Desktop config file.
What benchmarks are planned for MCP Reasoner?
Planned benchmarks include MATH500, GPQA-Diamond, GMSK8, and possibly Polyglot or SWE-Bench.