Loading ToolBench
Multi-Agent Sequential Thinking MCP — multi-provider validation layer for sequential-thinking (Ollama/OpenAI/Anthropic/Gemini/Bedrock/GitHub/OpenRouter)
MAST-MCP defines 2 tools with complete JSON schemas and detailed descriptions. Both tools follow the upstream sequential-thinking contract. Schemas include proper type definitions, minimums, and required fields. Descriptions are comprehensive (194 - 250 chars, within baseline 34 - 392 range) and explain WHEN to use each tool. However, parameter descriptions lack actionable constraints (e.g., no guidance on thought length limits, no enum for debate modes in mast_debate). Error handling is absent, no recovery guidance or categorization. Tool composition is sound (each does one thing), but the mast_debate tool conflates multiple reasoning strategies (Critic+Judge, De Bono, Actor-Critic, etc.) into one tool with optional model overrides, which could confuse LLM selection. No output schema documented. Security: no secrets exposed as params (good), but no audit logging visible in provided code.
Forces a specific reasoning strategy (Critic+Judge debate, De Bono Six Hats, Actor-Critic, Brainstorm, Tree of Thoughts, Kalman Convergence, or Workflow) regardless of server defaults. Use for maximum validation quality on critical reasoning steps. In debate mode: thought evaluated by Critic (identifies flaws) and Judge (verdict + revision). In debono mode: 7 sequential De Bono hats refine a working document. In actor_critic mode: iterative Critic+Judge loop up to ACTOR_CRITIC_MAX_ROUNDS. In brainstorm mode: N parallel generators produce ideas, Synthesizer merges top picks. In tot mode (Tree of Thoughts): N parallel branch generators, Voter scores them. In kalman mode: N scorers evaluate quality, Kalman filter fuses scores optimally. In workflow mode: chains multiple modes in sequence defined by MAST_WORKFLOW_STAGES. Accepts optional model overrides per call. All other parameters are identical to the sequentialthinking tool.
A tool for dynamic and reflective problem-solving through thoughts. This tool helps analyze problems through a flexible thinking process that can adapt and evolve. Each thought can build on, question, or revise previous insights as understanding deepens. When to use this tool: - Breaking down complex problems into steps - Planning and design with room for revision - Analysis that might need course correction - Problems where the solution path isn't clear upfront - Problems that require a multi-step solution - Tasks that need careful consideration of multiple aspects Key features: - You can adjust total_thoughts up or down as you progress - You can question or revise previous thoughts - You can add more thoughts even after reaching what seemed like the end - You can express uncertainty and explore alternative approaches Parameters explained: - thought: Your current thinking step - next_thought_needed: True if you need more thinking, even if at what you thought was the end - thought_number: Current thought number - total_thoughts: Current estimate of thoughts needed (can be adjusted) - is_revision: A boolean indicating if this thought revises previous thinking - revises_thought: If is_revision is true, which thought number is being reconsidered - branch_from_thought: Branching point thought number - branch_id: Branch identifier - needs_more_thoughts: If you realize you need more thoughts You should: 1. Start with an initial estimate of needed thoughts 2. Feel free to question or revise previous thoughts 3. Don't hesitate to add more thoughts if needed 4. Use is_revision when reconsidering previous thinking 5. Only set next_thought_needed to false when truly done
mast_debate conflates 7 distinct reasoning strategies (Critic+Judge, De Bono, Actor-Critic, Brainstorm, Tree of Thoughts, Kalman, Workflow) into one tool with optional model overrides. LLMs cannot infer which strategy to use without explicit guidance. Should split into separate tools (debate_critic_judge, debate_debono, debate_actor_critic, etc.) or add a required 'strategy' enum parameter.
No output schema documented for either tool. LLMs cannot plan downstream calls or extract structured data. Must document what fields are returned (e.g., does sequentialthinking return the thought text, metadata, or a structured reasoning trace?).
Parameter descriptions lack actionable constraints. 'thought' param has no length limit guidance. 'criticModel' and 'judgeModel' accept free-form strings with only examples ('mistral:7b-instruct', 'deepseek-r1:8b'), no enum or validation pattern. LLMs will hallucinate invalid model names.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | <=2025-11-25 | v2 |
No error handling or recovery guidance. If a thought is malformed, a model override is invalid, or reasoning fails, the tool provides no actionable error message. LLMs cannot self-correct.
Tool names do not start with action verbs. 'sequentialthinking' and 'mast_debate' are nouns/adjectives. Should be 'think_sequentially' or 'reason_step_by_step' and 'debate_reasoning' or 'validate_reasoning'. Current names obscure intent.