MCP Think

# Example Usage of the MCP Think Tool This document demonstrates how to use the MCP Think tool effectively in various scenarios. ## Basic Setup First, install the package: ```bash pip install mcp-think ``` Then, you can run the server: ```bash mcp-think ``` ## Example 1: Complex Problem Solving When an LLM is working on a multi-step problem: ``` Human: Can you help me optimize our company's supply chain? We have 5 warehouses and 12 retail locations, and we need to minimize shipping costs while ensuring timely delivery. LLM: I'll help you optimize your supply chain. Let me think through this problem systematically. [LLM uses the think tool] Thought: This is a complex optimization problem with multiple variables to consider: Locations of 5 warehouses and 12 retail locations (need coordinates or distances) Shipping costs between each warehouse-retail pair Delivery time constraints for each retail location Inventory capacity at each warehouse Demand patterns at each retail location I need to: Gather the complete data set Formulate this as a linear programming problem Consider both cost minimization and time constraints Possibly implement a modified transportation algorithm Evaluate multiple solutions for robustness [LLM responds with a structured approach based on the thinking] ``` ## Example 2: Policy Adherence When an LLM needs to ensure compliance with complex policies: ``` Human: I want to return this product I bought 45 days ago, but I lost the receipt. The package is opened. Can I still get a refund? LLM: Let me check our return policy carefully. [LLM uses the think tool] Thought: I need to analyze this return request against our store policy: Item was purchased 45 days ago Customer has no receipt Package is opened According to standard policy, returns require: Returns within 30 days (this is outside that window at 45 days) Original receipt (customer doesn't have this) Unopened packaging for certain items (package is opened) Exceptions to consider: Store credit might be possible even without receipt if we can verify purchase in our system Manager override for the 45-day window might be possible Product category matters - electronics vs. clothing have different policies Loyalty program members may have extended return windows I need to explain these policy details and suggest possible options. [LLM responds with a policy-compliant answer] ``` ## Example 3: Tool Output Analysis When an LLM needs to process results from multiple tool calls: ``` Human: Search our customer database and find all premium users who haven't logged in for 60+ days, then draft an email campaign to re-engage them. LLM: I'll help you identify inactive premium users and create a re-engagement campaign. [LLM uses database search tool, gets results] [LLM uses the think tool] Thought: I've received the query results showing 342 premium users who haven't logged in for 60+ days. Let me analyze this data: Patterns I'm noticing: 45% of inactive users are in the 6-12 month subscription cohort Most common last activity was "viewed dashboard" (72%) Very few (8%) used our newest features before becoming inactive 28% had support tickets in their last active month For the email campaign, I should: Segment these users into 2-3 groups based on their usage patterns Focus on new features they haven't explored Address possible pain points for users who had support tickets Include personalized incentives based on their historical usage Create a clear, compelling call-to-action [LLM proceeds to design a targeted campaign based on this analysis] ``` ## Integration with MCP Configuration To use this tool with Claude in Windsurf, add the following configuration to your MCP config file: ```json "think": { "command": "/path/to/mcp-think", "args": [], "type": "stdio", "pollingInterval": 30000, "startupTimeout": 30000, "restartOnFailure": true } ``` The `command` field should point to the directory where you installed the python package using pip.