spences10/mcp-sequentialthinking-tools

npx -y mcp-sequentialthinking-tools (Claude Desktop / compatible MCP clients)

Usage

• •Call sequentialthinking_tools with thought, thought_number, total_thoughts, next_thought_needed; optional session_id, branch/revise params, available_tools and recommended_tools
• •get_thinking_history returns stored thoughts for a session (limit default 50, max 500)
• •clear_thinking_history clears one session or all sessions
• •A sequential-thinking-guidance prompt tells the model to use the server honestly as a scratchpad, not an external reasoning engine

Key features

• ✓Scratchpad with per-session history, branching, revisions, and remaining-steps tracking
• ✓Validates recommended_tools against a supplied available_tools list; returns isError: true and skips storage if a name is missing
• ✓Security posture: treats thought text, tool descriptions, rationales, and remaining-step text as untrusted input — prompt-injection-like text is scanned and redacted, with security_warnings on redacted calls
• ✓MAX_HISTORY_SIZE per session (default 1000)
• ✓stdio transport accepts both standard Content-Length framed MCP messages and newline-delimited JSON used by older tmcp tooling

Best for

Tasks that benefit from explicit, inspectable reasoning — breaking messy problems into steps, revising or branching plans, or keeping a small reasoning history per session.

Caveats

• ⚠Does not discover your other MCP tools and does not choose tools for the model
• ⚠Adds overhead — README says not to use it for trivial requests
• ⚠Redaction is defensive filtering, not a guarantee; do not put secrets in thoughts or tool descriptions