Taskmaster

/** * models.js * MCP tool for managing AI model configurations */ import { createErrorResponse, handleApiResult, withToolContext } from '@tm/mcp'; import { z } from 'zod'; import { modelsDirect } from '../core/task-master-core.js'; /** * Register the models tool with the MCP server * @param {Object} server - FastMCP server instance */ export function registerModelsTool(server) { server.addTool({ name: 'models', description: 'Get information about available AI models or set model configurations. Run without arguments to get the current model configuration and API key status for the selected model providers.', parameters: z.object({ setMain: z .string() .optional() .describe( 'Set the primary model for task generation/updates. Model provider API key is required in the MCP config ENV.' ), setResearch: z .string() .optional() .describe( 'Set the model for research-backed operations. Model provider API key is required in the MCP config ENV.' ), setFallback: z .string() .optional() .describe( 'Set the model to use if the primary fails. Model provider API key is required in the MCP config ENV.' ), listAvailableModels: z .boolean() .optional() .describe( 'List all available models not currently in use. Input/output costs values are in dollars (3 is $3.00).' ), projectRoot: z .string() .describe('The directory of the project. Must be an absolute path.'), openrouter: z .boolean() .optional() .describe('Indicates the set model ID is a custom OpenRouter model.'), ollama: z .boolean() .optional() .describe('Indicates the set model ID is a custom Ollama model.'), bedrock: z .boolean() .optional() .describe('Indicates the set model ID is a custom AWS Bedrock model.'), azure: z .boolean() .optional() .describe('Indicates the set model ID is a custom Azure OpenAI model.'), vertex: z .boolean() .optional() .describe( 'Indicates the set model ID is a custom Google Vertex AI model.' ), 'openai-compatible': z .boolean() .optional() .describe( 'Indicates the set model ID is a custom OpenAI-compatible model. Requires baseURL parameter.' ), baseURL: z .string() .optional() .describe( 'Custom base URL for providers that support it (e.g., https://api.example.com/v1).' ) }), execute: withToolContext('models', async (args, context) => { try { context.log.info( `Starting models tool with args: ${JSON.stringify(args)}` ); // Use args.projectRoot directly (normalized by withToolContext) const result = await modelsDirect( { ...args, projectRoot: args.projectRoot }, context.log, { session: context.session } ); return handleApiResult({ result, log: context.log, errorPrefix: 'Error managing models', projectRoot: args.projectRoot }); } catch (error) { context.log.error(`Error in models tool: ${error.message}`); return createErrorResponse(error.message); } }) }); }