概要
What is Structured Thinking MCP Server?
A TypeScript Model Context Protocol (MCP) server that allows LLMs to programmatically construct mind maps for exploring an idea space, with enforced metacognitive self-reflection. It is based on Arben Ademi's Sequential Thinking Python server.
How to use Structured Thinking MCP Server?
Configure the tool in Claude Desktop, Cursor, or another MCP client using the JSON setting with "command": "npx" and "args": ["-y", "structured-thinking"]. The server exposes MCP tools such as capture_thought, revise_thought, retrieve_relevant_thoughts, get_thinking_summary, and clear_thinking_history.
Key features of Structured Thinking MCP Server
- Thought quality scores (0–1) for metacognitive feedback
- Thought stages (e.g., Problem Definition, Analysis) to steer thinking
- Branching to explore parallel lines of reasoning
- Short-term memory buffer (10 most recent thoughts)
- Long-term memory retrieval based on tags
- Summarization of entire thinking process
Use cases of Structured Thinking MCP Server
- LLMs constructing structured mind maps during problem-solving
- Enforcing self-reflection and metacognitive awareness in reasoning
- Exploring multiple solution branches simultaneously
- Debugging or reviewing an LLM's thought process step-by-step
FAQ from Structured Thinking MCP Server
How are thought quality scores used?
Each thought is assigned a quality score between 0 and 1, which combines with the thought's stage to provide metacognitive feedback that steers the LLM's thinking process.
What are thought stages and why are they important?
Each thought is tagged with a stage (e.g., Problem Definition, Analysis, Ideation). If the LLM spends too long in a stage or has low-quality thoughts, the server gives feedback to steer toward other stages or different thinking strategies.
Does the server persist thoughts to a file or database?
No. Currently, all thoughts are stored only in memory. There is no user interface or persistence to a file or database.
What are the main limitations of the server?
The metacognitive monitoring is mechanical, based on a single self-reported quality score and naive stage-based multipliers. There is also no user interface or visualization of the thought graph.
What MCP tools does the server expose?
It exposes capture_thought, revise_thought, retrieve_relevant_thoughts, get_thinking_summary, and clear_thinking_history.
