Yak MCP

import { FastMCP } from "fastmcp"; import { z } from "zod"; import type { Config } from "./config"; type MCPServerConfig = { name: string; version: `${number}.${number}.${number}`; onSay: (text: string, config: Config["tts"]) => void; }; export class MCPServer { private name: string; private version: `${number}.${number}.${number}`; private onSay: (text: string, config: Config["tts"]) => void; private server: FastMCP | undefined; constructor(config: MCPServerConfig) { this.name = config.name; this.version = config.version; this.onSay = config.onSay; } async start(config: Config) { if (this.server) { await this.server.stop(); this.server = undefined; } this.server = new FastMCP({ name: this.name, version: this.version, }); this.server.addTool({ name: "say", description: `WHEN TO USE: Use this tool when the coding agent needs to speak directly to a human — for reporting important events, announcing task completion, expressing emotions, or casually murmuring about less critical actions. HOW TO USE: - Provide the "text" parameter with the message to deliver. - Choose the "volume" parameter according to the situation: - SuperShouting: Critical alerts, long-running task completion, or urgent reports. - Shouting: Celebrations of success or expressions of sadness over failures. - Normal: Routine updates on upcoming or ongoing work. - Whisper: Informal comments, minor updates, or unimportant side notes. FEATURES: - Allows the coding agent to convey not only information but also tone and emotion. - Supports four distinct speaking styles to match the context. LIMITATIONS: - Does not process or interpret human speech; output only. - Emotional nuance is limited to the predefined volume levels. - Overuse of SuperShouting or Shouting may overwhelm the human listener. TIPS: - Reserve SuperShouting for truly critical or celebratory events to maintain its impact. - Use Shouting strategically to highlight both positive and negative outcomes. - Employ Normal for steady narration of workflow. - Leverage Whisper for humor, side comments, or low-priority notes to keep interactions natural. `, parameters: z.object({ text: z.string().describe(`The text to be spoken aloud, provided in ${config.tts.lang}.`), volume: z .enum(["SuperShouting", "Shouting", "Normal", "Whisper"]) .describe("The speaking volume level, ranging from very loud (SuperShouting) to very quiet (Whisper)."), }), execute: async (args) => { const volume = args.volume === "SuperShouting" ? 1.0 : args.volume === "Shouting" ? 0.9 : args.volume === "Normal" ? 0.75 : 0.5; this.onSay(args.text, { ...config.tts, volume: config.tts.volume * volume }); }, }); await this.server.start({ transportType: "httpStream", httpStream: { host: config.mcp.host, port: config.mcp.port, stateless: true, }, }); } async stop() { if (this.server) { await this.server.stop(); this.server = undefined; } } }