OpenRouter MCP Server

Instructions

Implementation Reference

• src/server.ts:330-361 (handler)

  The main handler function that executes the chat_with_model tool. It validates parameters using ChatRequestSchema, constructs the messages array (including optional system prompt), makes an API call to OpenRouter's chat/completions endpoint, and returns the response with usage statistics.

  private async chatWithModel(params: z.infer<typeof ChatRequestSchema>) {
    const { model, message, max_tokens, temperature, system_prompt } = params;
  
    const messages = [];
    if (system_prompt) {
      messages.push({ role: "system", content: system_prompt });
    }
    messages.push({ role: "user", content: message });
  
    const response = await axios.post(
      `${OPENROUTER_CONFIG.baseURL}/chat/completions`,
      {
        model,
        messages,
        max_tokens,
        temperature,
      },
      { headers: OPENROUTER_CONFIG.headers }
    );
  
    const result = response.data.choices[0].message.content;
    const usage = response.data.usage;
  
    return {
      content: [
        {
          type: "text" as const,
          text: `**Model:** ${model}\n**Response:** ${result}\n\n**Usage:**\n- Prompt tokens: ${usage.prompt_tokens}\n- Completion tokens: ${usage.completion_tokens}\n- Total tokens: ${usage.total_tokens}`,
        },
      ],
    };
  }

• src/server.ts:19-25 (schema)

  Zod schema defining the input validation for chat_with_model tool. Includes required fields (model, message) and optional fields (max_tokens, temperature, system_prompt) with defaults.

  const ChatRequestSchema = z.object({
    model: z.string().describe("OpenRouter model ID (e.g., 'openai/gpt-4')"),
    message: z.string().describe("Message to send to the model"),
    max_tokens: z.number().optional().default(1000).describe("Maximum tokens in response"),
    temperature: z.number().optional().default(0.7).describe("Temperature for response randomness"),
    system_prompt: z.string().optional().describe("System prompt for the conversation"),
  });

• src/server.ts:145-176 (registration)

  Tool registration in the ListToolsRequestSchema handler. Defines the tool's metadata, input schema structure, required parameters, and default values for MCP clients to discover.

  {
    name: "chat_with_model",
    description: "Send a message to a specific OpenRouter model",
    inputSchema: {
      type: "object",
      properties: {
        model: {
          type: "string",
          description: "OpenRouter model ID (e.g., 'openai/gpt-4')",
        },
        message: {
          type: "string",
          description: "Message to send to the model",
        },
        max_tokens: {
          type: "number",
          description: "Maximum tokens in response",
          default: 1000,
        },
        temperature: {
          type: "number",
          description: "Temperature for response randomness",
          default: 0.7,
        },
        system_prompt: {
          type: "string",
          description: "System prompt for the conversation",
        },
      },
      required: ["model", "message"],
    },
  },

• src/server.ts:229-230 (registration)

  Tool dispatch logic in the CallToolRequestSchema handler. Routes the chat_with_model tool invocation to the chatWithModel handler method with parameter validation.

  case "chat_with_model":
    return await this.chatWithModel(ChatRequestSchema.parse(args));

• src/server.ts:34-43 (helper)

  Configuration object used by the chatWithModel handler for API requests to OpenRouter. Contains base URL, API key, and headers required for authentication.

  const OPENROUTER_CONFIG = {
    baseURL: process.env.OPENROUTER_BASE_URL || "https://openrouter.ai/api/v1",
    apiKey: process.env.OPENROUTER_API_KEY,
    headers: {
      "Authorization": `Bearer ${process.env.OPENROUTER_API_KEY}`,
      "HTTP-Referer": process.env.OPENROUTER_SITE_URL || "http://localhost:3000",
      "X-Title": process.env.OPENROUTER_APP_NAME || "OpenRouter MCP Server",
      "Content-Type": "application/json",
    },
  };