Sequential Thinking

Sequential Thinking

Break down complex work into ordered steps, revise when needed, and produce one clear answer. Use for multi-step tasks, planning, and problems where scope or approach may change.

When to Use

• Complex or multi-step tasks
• Planning and design with room for revision
• Analysis that might need course correction
• Problems where the full scope is unclear at the start
• Tasks that need context carried across steps
• When irrelevant information must be filtered out
• When the user asks for step-by-step reasoning

Behavior Without the MCP Tool

1. Break the task into ordered steps.
2. State one step at a time; build on the previous step.
3. Revise earlier steps when you discover they are wrong or incomplete.
4. Branch or backtrack when exploring alternatives.
5. Generate a solution hypothesis when appropriate; verify it against the steps.
6. Repeat until satisfied, then give a single, clear final answer.

Principles: Start with an estimate of how many steps you need and adjust as you go. Express uncertainty. Ignore information that is irrelevant to the current step.

Using the Sequential Thinking MCP Tool

When the sequentialthinking MCP tool is enabled (e.g. user-sequential-thinking server), prefer calling it for hard problems so thoughts are tracked and revisable.

• Required every call: thought (current step), thoughtNumber, totalThoughts, nextThoughtNeeded.
• Start with an initial estimate of totalThoughts; increase or decrease as you progress.
• Set nextThoughtNeeded: true until you are done; set nextThoughtNeeded: false only when you have a satisfactory final answer.
• Revisions: Set isRevision: true and revisesThought to the thought number you are reconsidering.
• Branches: Use branchFromThought (thought number) and branchId (e.g. "alt-1") when exploring a different path.
• More thoughts later: Set needsMoreThoughts: true when you reach what seemed like the end but realize you need more steps.

Use the tool to generate a hypothesis, verify it against the chain of thought, and repeat until satisfied before returning one correct answer.

Principles (With or Without Tool)

• Start with an estimate of steps; adjust freely.
• Revise or add steps at any time.
• Express uncertainty when present.
• Ignore irrelevant information for the current step.
• Hypothesize and verify when useful.
• End with a single, clear answer when satisfied.