From sequential-thinking
Use this skill when you need to break down complex problems, plan multi-step implementations, analyze tricky bugs, make architectural decisions, reason through ambiguous requirements, or any task that benefits from structured step-by-step thinking. Also use when you need to revise earlier assumptions, explore alternative approaches, or verify a hypothesis before committing to an implementation.
How this skill is triggered — by the user, by Claude, or both
Slash command
/sequential-thinking:sequential-thinkingThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
The sequential-thinking MCP server provides a tool for dynamic, reflective
The sequential-thinking MCP server provides a tool for dynamic, reflective problem-solving through structured thought steps. Unlike simple chain-of-thought, it supports revision, branching, and adaptive depth.
mcp__sequential-thinking__sequentialthinkingEach call represents one thought step. Key parameters:
| Parameter | Type | Description |
|---|---|---|
thought | string | Your current thinking step |
thoughtNumber | int | Current step number (1-based) |
totalThoughts | int | Estimated total steps (adjustable) |
nextThoughtNeeded | bool | true if more thinking needed |
isRevision | bool | Whether this revises a prior thought |
revisesThought | int | Which thought number is being revised |
branchFromThought | int | Which thought to branch from |
branchId | string | Identifier for the branch |
needsMoreThoughts | bool | Signal that you need more steps |
For straightforward problems where each step builds on the last:
Thought 1: Understand the problem space
Thought 2: Identify key constraints
Thought 3: Propose solution approach
Thought 4: Verify approach against constraints
Thought 5: Final answer (nextThoughtNeeded: false)
When early assumptions prove wrong:
Thought 1: Initial hypothesis
Thought 2: Gather evidence
Thought 3: Evidence contradicts hypothesis
Thought 4: (isRevision=true, revisesThought=1) Revised hypothesis
Thought 5: Verify revised hypothesis
When multiple approaches deserve evaluation:
Thought 1: Problem statement
Thought 2: Approach A analysis
Thought 3: (branchFromThought=1, branchId="B") Approach B analysis
Thought 4: Compare branches
Thought 5: Select best approach
Start with a small estimate, expand as needed:
Thought 1/3: Initial analysis
Thought 2/3: Deeper than expected...
Thought 3/5: (needsMoreThoughts=true, totalThoughts=5) Need more analysis
Thought 4/5: Additional investigation
Thought 5/5: Conclusion
nextThoughtNeeded: falseThought 1: Identify the symptoms (intermittent failure, timing-dependent)
Thought 2: Map the concurrent operations involved
Thought 3: Identify shared state access points
Thought 4: Hypothesize: shared counter accessed without lock
Thought 5: Verify: trace code paths to confirm hypothesis
Thought 6: Propose fix with verification strategy
Thought 1: Document access patterns and query requirements
Thought 2: Branch A - normalized relational schema
Thought 3: Branch B - denormalized for read performance
Thought 4: Compare against access patterns
Thought 5: Revise - hybrid approach combining strengths
Thought 6: Validate against edge cases
Thought 1: Understand current architecture and pain points
Thought 2: Identify dependencies and blast radius
Thought 3: Propose incremental migration strategy
Thought 4: Identify risks and rollback plan
Thought 5: Break into atomic, shippable steps
This plugin:
.mcp.json (auto-registered on install)mcp__sequential-thinking__* to permissions.allow on session startnpx claudepluginhub nsheaps/ai-mktpl --plugin sequential-thinkingEnables structured step-by-step reasoning with revision and branching for complex problems, multi-stage analysis, design planning, problem decomposition, unclear scope, or alternative approaches.
Guides Claude through step-by-step analysis for complex problems using Sequential Thinking MCP. Activate with 'think step by step', 'reason through this', etc.
Externalizes stepwise reasoning as numbered, revisable, branchable thought blocks for multi-constraint problems, architecture decisions, and ambiguous bugs.