OpenRouter MCP Server

Instructions

Implementation Reference

• src/server.ts:363-413 (handler)

  The main implementation of compareModels tool that sends the same message to multiple models in parallel and formats the comparison results

  private async compareModels(params: z.infer<typeof CompareModelsSchema>) {
    const { models, message, max_tokens } = params;
  
    const promises = models.map(async (model) => {
      try {
        const response = await axios.post(
          `${OPENROUTER_CONFIG.baseURL}/chat/completions`,
          {
            model,
            messages: [{ role: "user", content: message }],
            max_tokens,
          },
          { headers: OPENROUTER_CONFIG.headers }
        );
  
        return {
          model,
          response: response.data.choices[0].message.content,
          usage: response.data.usage,
          success: true,
        };
      } catch (error) {
        return {
          model,
          error: error instanceof Error ? error.message : "Unknown error",
          success: false,
        };
      }
    });
  
    const results = await Promise.all(promises);
  
    const formattedResults = results
      .map((result) => {
        if (result.success) {
          return `**${result.model}:**\n${result.response}\n*Tokens: ${result.usage.total_tokens}*`;
        } else {
          return `**${result.model}:** ❌ Error - ${result.error}`;
        }
      })
      .join("\n\n---\n\n");
  
    return {
      content: [
        {
          type: "text" as const,
          text: `Comparison of ${models.length} models:\n\n${formattedResults}`,
        },
      ],
    };
  }

• src/server.ts:27-31 (schema)

  Zod schema defining the input validation for compare_models tool (models array, message, and optional max_tokens)

  const CompareModelsSchema = z.object({
    models: z.array(z.string()).describe("Array of model IDs to compare"),
    message: z.string().describe("Message to send to all models"),
    max_tokens: z.number().optional().default(500).describe("Maximum tokens per response"),
  });

• src/server.ts:177-202 (registration)

  Tool registration in the MCP tools list with name 'compare_models', description, and input schema

  {
    name: "compare_models",
    description: "Compare responses from multiple models",
    inputSchema: {
      type: "object",
      properties: {
        models: {
          type: "array",
          items: {
            type: "string",
          },
          description: "Array of model IDs to compare",
        },
        message: {
          type: "string",
          description: "Message to send to all models",
        },
        max_tokens: {
          type: "number",
          description: "Maximum tokens per response",
          default: 500,
        },
      },
      required: ["models", "message"],
    },
  },

• src/server.ts:231-232 (registration)

  Switch case handler that routes compare_models tool calls to the compareModels method

  case "compare_models":
    return await this.compareModels(CompareModelsSchema.parse(args));