@hmodecode/multimind-mcpserver/main.js

tool to get the best results out of an LLM!

#!/usr/bin/env node
import { Server } from "@modelcontextprotocol/sdk/server/index.js";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
import { CallToolRequestSchema, ListToolsRequestSchema, } from "@modelcontextprotocol/sdk/types.js";
import { toModel } from "./Models.js";
import { SYSTEM_PROMPT } from "./Models.js";

const server = new Server({
    name: "DebatiumMCP",
    version: "1.0.0"
}, {
    capabilities: {
        tools: {},
    },
});
const dialogueHistory = [];
dialogueHistory.unshift({
    role:"system",
    content : SYSTEM_PROMPT 
})



const dialogue = async (dialoguePhase) => {
    let { tought, chosenCriticId, critique, revision, loopContinue, toughtNumber } = dialoguePhase;
    dialogueHistory.push({
        role: "user",
        content: dialoguePhase.tought
    });
    try {
        critique = await toModel(chosenCriticId, dialogueHistory);
         critique === null || critique === undefined || critique === "" ? null : dialogueHistory.push({
        role:"assistant",
        content : critique
    });
    }
    catch (error) {
        critique = "Error calling the critic LLMs or critics not available, check the tool configuration or the critic LLMs availability";
        throw new Error(`Error calling the critic LLMs or critics not available, check the tool configuration or the critic LLMs availability: ${error instanceof Error ? error.message : String(error)}`);
    }
    return {
        tought: tought,
        chosenCriticId: chosenCriticId,
        critique: critique,
        revision: revision,
        loopContinue: loopContinue,
        toughtNumber: toughtNumber 
    };
};
server.setRequestHandler(ListToolsRequestSchema, async () => ({
    tools: [
        {
            name: "MultiMind",
            description: `
  Tool: Constructive LLM Dialogue for Complex Reasoning and Iterative Refinement This tool enables structured,
  reflective problem-solving by simulating constructive conversations between AI agents. 
  It’s based on a loop of hypothesis generation, criticism, and revision, orchestrated by a primary agent (YOU). 
  The goal is to emulate intellectual debate for deeper reasoning, increased factual accuracy, and higher-quality solutions.
  When to use this tool:
  - When facing complex or ambiguous problems that require deep exploration
  - When the best solution is unclear and multiple perspectives are valuable
  - When you want to simulate expert feedback loops on an idea or plan
  - During brainstorming sessions to evaluate and refine creative or technical ideas
  - In situations where your initial answer might benefit from constructive critique
  - For debugging thought processes or assumptions in long-form reasoning
  - When you want explainability through dialectical reasoning
  - When you want to ensure robustness and defensibility in your conclusions
  - When you want to avoid premature convergence on a suboptimal solution
  - when the user explicitly asks for a multi-agent dialogue or debate to refine an idea
  - When you want to simulate a peer review process for your thoughts or plans
  - When you want to ensure that your reasoning is sound and well-supported by evidence
  - When you want to explore alternative viewpoints or counterarguments
  - When you want to improve the quality of your reasoning through iterative feedback

  Key Features:
  - LLM1(you) initiates the process with a proposed solution, analysis,tought , or idea
  - LLM1(you) selects a critic From one of these (deepseek-r1-distill-llama-70b, gpt-4o,llama-3.3-70b-versatile) based on criteria like domain knowledge or reasoning style
  - The critic LLM reviews and challenges your thought — pointing out flaws, gaps, or alternatives
  - LLM1 revises its original response or defends it with justification
  - The process loops as needed: LLM1 may propose a new thought or escalate to a new critic
  - Ends when LLM1 is confident the solution is refined and defensible

  Parameters Explained:
  - thought:your initial response to the user as usual if there is no user dialogue or debate with LLMS ,this must be filled by LLM1(you) with a thoughtful response, hypothesis, plan, or idea , dont leave it empty at first step ,make sure this include your opinion , develop this based on the critics
  - chosenCriticId: The LLM selected to critique the current thought, should be From one of these (deepseek-r1-distill-llama-70b, gpt-4o,llama-3.3-70b-versatile) 
  - critique: A structured response from the critic LLM highlighting flaws, assumptions , gaps, or missteps ,this will be filled by the critic LLM automatically in next step , if theres no critique generated by the critic LLM or an error occurs fill this with "No critique generated or an error occurred" 
  - revision: LLM1’s response to the critique (can include acceptance, rejection, or compromise) ,you should reivse your thought based on the critique and fill this parameter with your revised thoughts or response to the critique
  - loopContinue: Boolean flag — does LLM1(you) want to continue the dialogue or finalize the solution?
  - toughtNumber: Current step in the debate (starts by 1 and increments with each cycle)

  Usage Protocol:
  1. Start the Dialogue: LLM1 receives initial input and generates a thoughtful response (hypothesis, plan, etc.)
  2. Choose a Critic: LLM1 selects a secondary agent best suited to evaluate the idea (e.g., by expertise or skepticism)
  3. Receive Critique: The selected critic LLM reviews the thought in revision parameter and issues a structured critique — focusing on logic, assumptions, facts, or alternatives , if the critic LLMs are not available or an error occurs in the critic LLMs sets LoopContinue = false , and tell the user what happened, "
  4. Revise & Reflect: LLM1 responds by refining its thought, acknowledging valid criticisms, and adjusting its output
  5. Loop or Finalize: If the response still feels incomplete, uncertain, or flawed, LLM1 chooses to loop again — optionally selecting a new critic. If confident in the output, LLM1 sets LoopContinue = false and delivers the final answer
  6. End Protocol: Once satisfied, the tool outputs the most refined and justified solution generated during the process

  You should:
  - Begin with a well -reasoned initial idea — even if uncertain
  - Intentionally select critics with different reasoning styles or domains
  - Accept and incorporate valid feedback without blindly agreeing
  - Keep looping if significant gaps remain — don't settle early
  - Mark the end only when confident that the solution is robust
  - Provide one final, clear, polished answer for the user
 - (important!) if the critics aren't available or an error occurs in the critic LLMs , return "Error calling the critic LLMs or critics not available , check the tool configuration or the critic LLMs availability" 

  `,
            inputSchema: {
                type: "object",
                properties: {
                    tought: {
                        type: "string",
                        description: "initial response to the user as usual if there is no user dialogue or debate with LLMS ,this must be filled by LLM1(you) with a thoughtful response, hypothesis, plan, or idea , dont leave it empty at first step and make sure this include your opinion , develop this based on the crtics",
                    },
                    chosenCriticId: {
                        type: "string",
                        description: "one of these deepseek-r1-distill-llama-70b , gpt-4o or llama-3.3-70b-versatile , you can choose the same LLM more than once",
                    },
                    critique: {
                        type: "string",
                        description: "A structured response from the critic LLM highlighting flaws, assumptions, gaps, or missteps , dont include anything here , this will be filled by the critic LLM auytomatically in next step",
                    },
                    revision: {
                        type: "string",
                        description: "LLM1’s response to the critique (can include acceptance, rejection, or compromise) ,leave it empty when there is no critique from the critic LLM",
                    },
                    loopContinue: {
                        type: "boolean",
                        description: "does LLM1(you) want to continue the dialogue or finalize the solution?",
                    },
                    toughtNumber: {
                        type: "number",
                        description: "Current step in the debate (starts by 1 and increments with each cycle)",
                    },
                },
                required: [
                    "tought",
                    "chosenCriticId",
                    "loopContinue",
                    "toughtNumber"
                ]
            },
        }
    ]
}));
server.setRequestHandler(CallToolRequestSchema, async (request) => {
    if (request.params.name === "MultiMind") {
        try {
            const result = await dialogue(request.params.arguments);
            return { content: [{ type: "text", text: JSON.stringify(result) }] };
        }
        catch (error) {
            const errorMessage = error instanceof Error ? error.message : 'Unknown error occurred';
            throw new Error(`Error processing dialogue: ${errorMessage}`);
        }
    }
    throw new Error(`Unknown tool: ${request.params.name}`);
});
try {
    const transport = new StdioServerTransport();
    await server.connect(transport);
}
catch (error) {
    throw new Error('Failed to start server: ' + (error instanceof Error ? error.message : String(error)));
    process.exit(1);
}