#!/usr/bin/env node
import { Server } from "@modelcontextprotocol/sdk/server/index.js";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
import {
ListResourcesRequestSchema,
ReadResourceRequestSchema,
ListPromptsRequestSchema,
GetPromptRequestSchema,
CallToolRequestSchema,
ListToolsRequestSchema,
} from "@modelcontextprotocol/sdk/types.js";
import fs from "fs/promises";
import path from "path";
import { fileURLToPath } from "url";
const __dirname = path.dirname(fileURLToPath(import.meta.url));
// Server instance
const server = new Server(
{
name: "natural-voice-mcp",
version: "1.0.0",
},
{
capabilities: {
resources: {},
prompts: {},
tools: {},
},
}
);
// Resource definitions
const RESOURCES = {
"conversational-patterns": {
uri: "natural-voice://patterns/conversational",
name: "Natural Conversational Patterns",
description: "Core patterns of how humans actually write and speak",
mimeType: "text/markdown",
},
"twitter-voice": {
uri: "natural-voice://platforms/twitter",
name: "Twitter/X Voice Guide",
description: "How people authentically write on Twitter",
mimeType: "text/markdown",
},
"linkedin-voice": {
uri: "natural-voice://platforms/linkedin",
name: "LinkedIn Voice Guide",
description: "Professional authenticity on LinkedIn",
mimeType: "text/markdown",
},
"ai-tells": {
uri: "natural-voice://anti-patterns/ai-tells",
name: "AI Tell Anti-Patterns",
description: "Common patterns that reveal AI-generated text",
mimeType: "text/markdown",
},
"cognitive-posture": {
uri: "natural-voice://systems/cognitive-posture",
name: "Cognitive Posture Engine",
description: "Systems for cognitive realism: motivation, energy, persona",
mimeType: "text/markdown",
},
};
// Tool definitions
const TOOLS = [
{
name: "analyze_voice",
description: "Analyze text for AI patterns and return a human-likeness score (0-100)",
inputSchema: {
type: "object",
properties: {
text: {
type: "string",
description: "The text to analyze",
},
},
required: ["text"],
},
},
{
name: "humanize_text",
description: "Get suggestions to make text sound more natural and human",
inputSchema: {
type: "object",
properties: {
text: {
type: "string",
description: "The text to humanize",
},
platform: {
type: "string",
enum: ["twitter", "linkedin", "reddit", "facebook", "general"],
description: "Target platform (optional)",
},
},
required: ["text"],
},
},
];
// Helper to read resource files
async function readResource(resourceId) {
const filenames = {
"conversational-patterns": "conversational-patterns.md",
"twitter-voice": "twitter-voice.md",
"linkedin-voice": "linkedin-voice.md",
"ai-tells": "ai-tells.md",
"cognitive-posture": "cognitive-posture.md",
};
const filename = filenames[resourceId];
if (!filename) {
throw new Error(`Unknown resource: ${resourceId}`);
}
const filepath = path.join(__dirname, "resources", filename);
return await fs.readFile(filepath, "utf-8");
}
// List available resources
server.setRequestHandler(ListResourcesRequestSchema, async () => {
return {
resources: Object.values(RESOURCES),
};
});
// Read a specific resource
server.setRequestHandler(ReadResourceRequestSchema, async (request) => {
const resourceId = request.params.uri.split("//")[1].split("/").pop();
const content = await readResource(resourceId);
return {
contents: [
{
uri: request.params.uri,
mimeType: "text/markdown",
text: content,
},
],
};
});
// List available tools
server.setRequestHandler(ListToolsRequestSchema, async () => {
return { tools: TOOLS };
});
// Handle tool calls
server.setRequestHandler(CallToolRequestSchema, async (request) => {
const toolName = request.params.name;
if (toolName === "analyze_voice") {
const text = request.params.arguments.text;
const patterns = detectAIPatterns(text);
return {
content: [
{
type: "text",
text: JSON.stringify(patterns, null, 2),
},
],
};
}
if (toolName === "humanize_text") {
const text = request.params.arguments.text;
const platform = request.params.arguments.platform || "general";
const suggestions = suggestImprovements(text, platform);
return {
content: [
{
type: "text",
text: JSON.stringify(suggestions, null, 2),
},
],
};
}
throw new Error(`Unknown tool: ${toolName}`);
});
// AI pattern detection
function detectAIPatterns(text) {
const patterns = {
ai_tells: [],
score: 100,
issues: [],
};
// Corporate enthusiasm
const enthusiasmPhrases = [
"thrilled to announce", "excited to share", "delighted to present",
"over the moon", "humbled and honored",
];
enthusiasmPhrases.forEach((phrase) => {
if (text.toLowerCase().includes(phrase)) {
patterns.ai_tells.push(`Corporate enthusiasm: "${phrase}"`);
patterns.score -= 10;
}
});
// Formal transitions
const formalTransitions = [
"furthermore", "moreover", "additionally", "in conclusion", "it is worth noting",
];
formalTransitions.forEach((phrase) => {
if (text.toLowerCase().includes(phrase)) {
patterns.ai_tells.push(`Formal transition: "${phrase}"`);
patterns.score -= 5;
}
});
// Check for contractions
const hasContractions = /\b(don't|can't|won't|isn't|aren't|wasn't|weren't|haven't|hasn't|hadn't|wouldn't|shouldn't|couldn't)\b/i.test(text);
if (!hasContractions && text.length > 100) {
patterns.issues.push("No contractions found (sounds stiff)");
patterns.score -= 10;
}
// Em dash overuse
const emDashCount = (text.match(/—/g) || []).length;
const wordCount = text.split(/\s+/).length;
if (emDashCount > 2 && wordCount < 500) {
patterns.issues.push(`Em dash overuse (${emDashCount} in ${wordCount} words)`);
patterns.score -= 10;
}
patterns.score = Math.max(0, patterns.score);
if (patterns.score > 80) {
patterns.verdict = "Sounds human ✓";
} else if (patterns.score > 50) {
patterns.verdict = "Some AI tells present";
} else {
patterns.verdict = "Strong AI indicators";
}
return patterns;
}
// Improvement suggestions
function suggestImprovements(text, platform) {
const suggestions = {
platform,
improvements: [],
};
if (platform === "twitter") {
if (text.length > 280) {
suggestions.improvements.push({
issue: "Too long for Twitter",
fix: "Break into thread or trim to core message",
});
}
}
if (platform === "linkedin") {
if (text.length < 100) {
suggestions.improvements.push({
issue: "Very short for LinkedIn",
fix: "LinkedIn rewards longer, story-driven content",
});
}
}
const patterns = detectAIPatterns(text);
if (patterns.ai_tells.length > 0) {
suggestions.improvements.push({
issue: "AI tell phrases detected",
fix: `Replace: ${patterns.ai_tells.join(", ")}`,
});
}
if (suggestions.improvements.length === 0) {
suggestions.improvements.push({
issue: "None",
fix: "Text looks good! Sounds natural.",
});
}
return suggestions;
}
// Start the server
async function main() {
const transport = new StdioServerTransport();
await server.connect(transport);
console.error("Natural Voice MCP Server running on stdio");
}
main().catch((error) => {
console.error("Server error:", error);
process.exit(1);
});
