LLM MCP Bridge

Instructions

Implementation Reference

• src/tools.ts:211-229 (handler)

  The core handler function for the 'llm_get_models' tool. It creates an LLMClient instance, lists available models from the server, and returns a JSON-formatted response with model IDs, owners, count, and baseURL.

  async llm_get_models(args: z.infer<typeof GetModelsSchema> = {}) {
    const client = getClient(args);
    const models = await client.listModels();
    return {
      content: [
        {
          type: "text" as const,
          text: JSON.stringify({
            models: models.map(m => ({
              id: m.id,
              owned_by: m.owned_by,
            })),
            count: models.length,
            baseURL: args.baseURL || defaultConfig.baseURL,
          }, null, 2),
        },
      ],
    };
  },

• src/tools.ts:72-82 (schema)

  MCP tool registration entry defining the name, description, and input schema (connection properties like baseURL and apiKey) for 'llm_get_models'.

  {
    name: "llm_get_models",
    description: "Obtiene la lista de modelos disponibles en el servidor LLM (compatible con OpenAI API: LM Studio, Ollama, vLLM, OpenAI, etc.)",
    inputSchema: {
      type: "object" as const,
      properties: {
        ...connectionProperties,
      },
      required: [],
    },
  },

• src/tools.ts:11-12 (schema)

  Zod schema for input validation of llm_get_models arguments, extending the base ConnectionConfigSchema.

  export const GetModelsSchema = ConnectionConfigSchema.extend({});

• src/index.ts:52-53 (registration)

  Registration in the MCP server's CallToolRequest handler that dispatches to the llm_get_models tool handler based on the tool name.

  case "llm_get_models":
    return await toolHandlers.llm_get_models(args as any);

• src/index.ts:42-44 (registration)

  MCP server handler for listing tools, which returns the tools array including 'llm_get_models'.

  server.setRequestHandler(ListToolsRequestSchema, async () => {
    return { tools };
  });