Reasoner - Advanced Multi-Strategy Reasoning

Core Concept

mcp__plugin_kg_kodegen__reasoner provides sophisticated reasoning with multiple search strategies. Unlike sequential_thinking (simple linear tracking), reasoner uses algorithms like Beam Search and Monte Carlo Tree Search (MCTS) to explore and score multiple solution paths, finding optimal reasoning chains.

Strategies

Strategy	Best For	Description
`beam_search`	General problems	Maintains top N paths simultaneously
`mcts`	Decision trees	UCB1/PUCT exploration-exploitation
`mcts_002_alpha`	Creative solutions	10% higher exploration bonus
`mcts_002alt_alpha`	Detailed analysis	Rewards longer reasoning paths

Key Parameters

Required:

Parameter	Type	Description
`thought`	string	Current reasoning step
`thought_number`	number	Current step (1-based)
`total_thoughts`	number	Estimated total needed
`next_thought_needed`	boolean	Whether more steps needed

Optional:

Parameter	Type	Description
`strategy_type`	string	beam_search (default), mcts, mcts_002_alpha, mcts_002alt_alpha
`beam_width`	number	Paths to maintain (1-10, default: 3)
`num_simulations`	number	MCTS rollouts (1-150, default: 50)
`parent_id`	string	Parent node for branching

Usage Examples

Beam Search (Default)

json

{
  "thought": "Analyzing possible caching strategies for the API",
  "thought_number": 1,
  "total_thoughts": 4,
  "next_thought_needed": true,
  "strategy_type": "beam_search",
  "beam_width": 3
}

MCTS for Decision Making

json

{
  "thought": "Evaluating database migration approaches",
  "thought_number": 1,
  "total_thoughts": 3,
  "next_thought_needed": true,
  "strategy_type": "mcts",
  "num_simulations": 100
}

Creative Problem Solving

json

{
  "thought": "Exploring novel approaches to distributed consensus",
  "thought_number": 1,
  "total_thoughts": 5,
  "next_thought_needed": true,
  "strategy_type": "mcts_002_alpha",
  "num_simulations": 75
}

Detailed Analysis

json

{
  "thought": "Deep comparison of microservices vs monolithic architecture",
  "thought_number": 1,
  "total_thoughts": 6,
  "next_thought_needed": true,
  "strategy_type": "mcts_002alt_alpha",
  "num_simulations": 50
}

Branching from Parent

json

{
  "thought": "Alternative approach using event sourcing",
  "thought_number": 3,
  "total_thoughts": 5,
  "next_thought_needed": true,
  "parent_id": "previous-node-uuid"
}

Output Format

json

{
  "session_id": "uuid-v4",
  "thought": "echoed input",
  "score": 0.85,
  "depth": 2,
  "is_complete": false,
  "next_thought_needed": true,
  "branches": 3,
  "best_path_score": 0.92,
  "strategy": "beam_search",
  "history_length": 5
}

When to Use What

Problem Type	Tool	Why
Simple step-by-step	sequential_thinking	No scoring needed
Optimization	reasoner (mcts)	Finds optimal path
Multiple alternatives	reasoner (beam_search)	Tracks top N paths
Creative exploration	reasoner (mcts_002_alpha)	Higher exploration
Detailed analysis	reasoner (mcts_002alt_alpha)	Rewards depth

Reasoner vs Sequential Thinking

Feature	Sequential Thinking	Reasoner
Path scoring	No	Yes (0.0-1.0)
Strategy selection	No	beam_search, MCTS variants
Semantic analysis	No	Yes (Stella 400M embeddings)
Best path tracking	No	Yes (best_path_score)
Complexity	Lower	Higher
Use case	Linear reasoning	Optimization/exploration

Remember

•Choose strategy wisely - beam_search for general, MCTS for optimization
•Adjust beam_width - higher = more paths but slower
•num_simulations - more = better MCTS results but slower
•Check scores - output includes quality scores (0.0-1.0)
•Use for complex problems - overkill for simple reasoning
•Prefer sequential_thinking for straightforward step-by-step