Sequential Thinking

Process

Phase 1: Problem Decomposition

# Check if session already loaded (server auto-loads last session on startup)
mcporter call sequential-thinking.analyze_session()

# If no active session, start one
mcporter call sequential-thinking.start_session(
  problem: "What fundamental questions must I answer?",
  success_criteria: "Clear understanding of core questions",
  session_type: "general"
)

Phase 2: Structured Discovery

# Add initial thought — returns thought_id needed for branching
mcporter call sequential-thinking.add_thought(
  content: "FIRST PRINCIPLE: What is this fundamentally?",
  confidence: 0.9
)
# → save the returned thought_id

# Fork into parallel reasoning tracks
mcporter call sequential-thinking.create_branch(
  name: "conceptual-analysis",
  from_thought: "<thought_id from above>",
  purpose: "Understand concepts separate from implementation"
)
# → save the returned branch_id

mcporter call sequential-thinking.add_thought(
  content: "Pattern criteria: Generic, clear interfaces, domain-independent",
  branch_id: "<branch_id from above>"
)

Phase 3: Synthesis

mcporter call sequential-thinking.merge_branch(
  branch_id: "<branch_id>",
  target_thought: ""
)

Phase 4: Pattern Extraction

mcporter call sequential-thinking.query_memories(
  tags: "domain-specific,architectural-patterns",
  content_contains: "core-concept"
)

mcporter call sequential-thinking.store_memory(
  content: "Brief description of the pattern",
  code_snippet: "pure code without comments",
  language: "python",
  tags: "pattern,category"
)

Full mcporter call reference

start_session

mcporter call sequential-thinking.start_session(
  problem: "what to solve",
  success_criteria: "how to know it is done",
  session_type: "general"        # or "coding" — triggers explore_packages automatically
)

Returns: session_id

add_thought

mcporter call sequential-thinking.add_thought(
  content: "reasoning step",
  branch_id: "<id>",             # optional — omit for main thread
  confidence: 0.8                # optional
)

Returns: thought_id — save this when you need to branch from it

create_branch

mcporter call sequential-thinking.create_branch(
  name: "branch-name",
  from_thought: "<thought_id>",  # required — must be non-empty string
  purpose: "why this branch exists"
)

Returns: branch_id

merge_branch

mcporter call sequential-thinking.merge_branch(
  branch_id: "<branch_id>",
  target_thought: ""             # optional — leave empty to merge into latest
)

Returns: branch_id, target_thought, status: "merged"

store_memory

mcporter call sequential-thinking.store_memory(
  content: "what to remember",
  code_snippet: "pure code",     # optional
  language: "python",            # optional
  tags: "tag1,tag2"              # optional — comma-separated
)

Returns: memory_id

query_memories

mcporter call sequential-thinking.query_memories(
  tags: "tag1,tag2",             # OR search; use & for AND: "tag1&tag2"
  content_contains: "/regex/"    # optional — wrap in // for regex
)

Returns: memories[], count, search_tips when empty

record_decision

mcporter call sequential-thinking.record_decision(
  decision_title: "What we decided",
  context: "why this decision was needed",
  options_considered: "option A vs option B",
  chosen_option: "option A",
  rationale: "because...",
  consequences: "this means..."
)

Returns: decision_id

explore_packages

mcporter call sequential-thinking.explore_packages(
  task_description: "what I am trying to do",
  language: "python"             # optional, default python
)

Returns: packages[], count

export_session

mcporter call sequential-thinking.export_session(
  filename: "output.md",
  format: "markdown",            # or "json"
  export_type: "session"         # or "memories"
)

list_sessions / load_session

mcporter call sequential-thinking.list_sessions()
mcporter call sequential-thinking.load_session(session_id: "<id>")

analyze_session

mcporter call sequential-thinking.analyze_session()

Returns stats or {"error": ...} if no active session — use this before start_session

Runtime Behaviors

Storage

memory-bank/
├── sessions/     # session .md files
├── memories/     # standalone memory .md files
└── patterns/     # pure code only (no comments, no docstrings)

Workflow Examples

Codebase Analysis

mcporter call sequential-thinking.start_session(
  problem: "Fundamental concepts in this codebase?",
  success_criteria: "Mental model of structure and reusable elements",
  session_type: "general"
)

mcporter call sequential-thinking.add_thought(content: "Initial observation about structure")
# save returned thought_id as t1

mcporter call sequential-thinking.create_branch(name: "architecture", from_thought: "<t1>", purpose: "Structural patterns")
mcporter call sequential-thinking.create_branch(name: "data-flow", from_thought: "<t1>", purpose: "Information flow")

mcporter call sequential-thinking.merge_branch(branch_id: "<arch-id>")
mcporter call sequential-thinking.merge_branch(branch_id: "<flow-id>")

Architecture Decision

mcporter call sequential-thinking.record_decision(
  decision_title: "Storage format choice",
  context: "Need persistent session storage",
  options_considered: "JSON vs Markdown",
  chosen_option: "Markdown",
  rationale: "Human-readable, diffable, no parser needed",
  consequences: "Regex parsing required for reload"
)

Anti-Patterns

• Calling start_session without checking analyze_session first — may create duplicate session
• Passing empty string to from_thought in create_branch — raises ValidationError
• Using dependencies param in add_thought — broken for non-empty values
• Tool-first thinking — define what to understand before calling tools
• Comments or docstrings in stored code patterns

Key Insight

Sequential thinking is NOT about organizing tools — it is about structuring COGNITION. Tools serve your thinking process, not replace it.