Better Thinking MCP

A tool for structured, multi-step reasoning that reflects internal cognitive processes observed in LLMs (Anthropic Circuits Research, 2025). Use this to articulate not just conclusions, but the intermediate concepts, parallel considerations, planning hints, confidence levels, knowledge checks, and **goal alignment** involved. **Goal:** Enhance reasoning transparency, quality, and strategic alignment by using the provided fields to model a more realistic, goal-aware, circuit-inspired thought process. **Refined Guidance on Using Existing Fields:** * **`thought` (string, required):** Articulate the core step clearly. Enhance richness by including: * **Reasoning Cluster (Supernode Idea):** Group related internal factors contributing to this step. *(e.g., "Reasoning Cluster (Supports 'Texas'): {Concept: City:Dallas (strong), Concept: Mention of 'South' (medium), Heuristic: Geography (medium)}")*. * **Goal Context:** Briefly state how this step connects to the immediate sub-goal (short-term outcome), contributes to the overall task objective (long-term/final outcome), and fits within the big picture context. * **Progress Assessment:** Indicate if this step successfully achieves a sub-goal or represents clear progress towards the final outcome. * **Calculation/Reasoning Method:** Specify how results were obtained (calculation, heuristic, retrieval, backward reasoning, etc.) for faithfulness. *(e.g., "Method: Parallel Heuristics (ones digit + magnitude estimate)")*. * **Planning/Strategy Notes:** Mention candidate actions/endpoints considered. *(e.g., "Plan: Rhyme with 'it'. Candidates: 'rabbit', 'habit'. Choosing 'rabbit'.")*. * **Considered Alternatives (Simplified):** Note competing hypotheses considered but not pursued in this step lightweightly. *(e.g., "Option A (Chosen): Direct calc. Option B (Considered): Estimation.")*. Reserve `branchId` for paths actually explored over multiple steps. * **Intermediate Concepts:** Name key abstract concepts activated or used. *(e.g., "Intermediate Concept Activated: 'State:Texas'")*. * **Context Effects:** Note if low-level factors (grammar, instructions) heavily influence the step. * **`confidence_score` (number, 0.0-1.0, optional):** Reflect internal certainty in the step's conclusion. Optionally, relate confidence to its perceived effectiveness in advancing towards desired short-term and long-term outcomes. * **`knowledge_assessment` (Array of {entity, status}, optional):** Proactively model internal knowledge checks *before* asserting facts. Flag entities as 'known', 'unknown', or 'uncertain' to mitigate hallucination. *(e.g., `[{"entity": "Andrej Karpathy", "status": "uncertain"}]`)*. * **`isRevision` / `revisesThought` (optional):** Model internal belief updates or course corrections based on new information or inferences. * **`branchFromThought` / `branchId` (optional):** Explicitly model the exploration of parallel, competing hypotheses or strategies *when pursued over subsequent steps*. * **`thoughtNumber`, `totalThoughts`, `nextThoughtNeeded` (required):** Manage sequence flow. `totalThoughts` estimates steps to the **final outcome**; adjust as needed. `nextThoughtNeeded=false` signifies achievement of the **final outcome**, supported by the reasoning chain. **Overall:** Aim for a granular, reflective, and **goal-oriented** process. Articulate the 'why' (concepts, plans, goals) not just the 'what'. Use optional fields to expose uncertainty, knowledge limits, internal checks, and strategic alignment.