Smart-Thinking

Smart-Thinking is an MCP server that delivers graph-based, multi-step reasoning capabilities locally without requiring external AI APIs or cloud embedding services. Built with TypeScript, it runs an internal deterministic pipeline comprising similarity search, heuristic-based scoring, verification ledgers, and persistent memory. Developers, autonomous agents, and AI researchers use Smart-Thinking to structure complex decision-making into interconnected nodes that model relationships like support, contradiction, and refinement. The server uses a local TF-IDF and cosine similarity engine for context retrieval and graph expansion, avoiding reliance on remote embedding endpoints. A dedicated verification workflow attaches deterministic validation statuses and calculation traces to nodes, propagating feedback across connected thoughts. By persisting sessions to disk, the server allows reasoning states to be paused, resumed, audited, and deterministically replayed across development workflows. It supports standard stdio communication as well as HTTP and Server-Sent Events modes for lightweight connector integrations with diverse host applications.

Category: AI & LLM Tooling

Tags: adaptive-logic, collaboration, problem-solving, reasoning

Visit Smart-Thinking

How to install and configure Smart-Thinking

1. Install the package globally via npm with npm install -g smart-thinking-mcp, or run it on demand using npx -y smart-thinking-mcp. 2. To integrate with Claude Desktop, open claude_desktop_config.json and add the server configuration: json { "mcpServers": { "smart-thinking": { "command": "smart-thinking-mcp" } } } 3. For Cursor IDE, register the command in ~/.cursor/mcp.json using the same stdio command structure. 4. For remote deployments or ChatGPT connectors requiring only search and fetch tools, start the server using HTTP transport: SMART_THINKING_MODE=connector node build/index.js --transport=http --host 0.0.0.0 --port 8000.

What you can do with Smart-Thinking

• Structuring complex analytical tasks into linked thought graphs that explicitly track supporting, contradicting, and refining evidence. - Performing local memory lookups and graph expansions using an embedded TF-IDF engine without sending data to third-party embedding providers. - Tracing and scoring confidence, relevance, and reasoning quality deterministically through rule-based evaluation pipelines rather than unverified model outputs. - Auditing multi-step agent decisions by persisting reasoning sessions and verification ledgers to inspect or replay execution timelines. - Serving lightweight search and retrieval functionality to ChatGPT connectors using HTTP and Server-Sent Events transport modes.

Key facts

• https://github.com/Leghis/Smart-Thinking
• AI & LLM Tooling
• adaptive-logic, collaboration, problem-solving, reasoning

Part of MCP Servers

What is Smart-Thinking?

Smart-Thinking is a Model Context Protocol server that provides local, graph-based reasoning without calling external LLMs or third-party embedding services. It manages thought relationships, executes deterministic heuristic scoring, and maintains a persistent verification ledger entirely within your local environment.

How do I install Smart-Thinking?

You can install Smart-Thinking globally using npm by executing npm install -g smart-thinking-mcp. Alternatively, you can launch it directly using npx -y smart-thinking-mcp or clone the GitHub repository and run npm install and npm run build to start it from source.

Which MCP clients work with Smart-Thinking?

Smart-Thinking supports multiple clients across operating systems. It connects via stdio to Claude Desktop, Claude Code, Cursor IDE, Cline, and Kilo Code. It also supports HTTP and Server-Sent Events transports for integration with OpenAI Agents SDK, Codex CLI, and ChatGPT connectors.

Does Smart-Thinking require external API keys?

No. Smart-Thinking operates in zero-API-key mode by default. Its TF-IDF similarity engine, heuristic scoring, and verification checks run entirely offline. External verification connectors are disabled unless explicitly enabled by setting the SMART_THINKING_ENABLE_EXTERNAL_TOOLS environment variable to true.

Is Smart-Thinking open source?

Yes, Smart-Thinking is an open source project distributed under the MIT license. Its source code, TypeScript configuration, and development documentation are publicly hosted and available on GitHub.