Swagger MCP

import dotenv from 'dotenv'; import minimist from 'minimist'; import { Server } from "@modelcontextprotocol/sdk/server/index.js"; import { StdioServerTransport } from '@modelcontextprotocol/sdk/server/stdio.js'; import { CallToolRequestSchema, ListToolsRequestSchema, ListPromptsRequestSchema, GetPromptRequestSchema } from "@modelcontextprotocol/sdk/types.js"; // Load environment variables from .env file dotenv.config(); import logger from "./utils/logger.js"; // Helper to sanitize log output (removes newlines and control chars) function sanitizeForLog(input: string): string { return input.replace(/[\r\n\x00-\x1F\x7F]+/g, ''); } // Parse command line arguments const argv = minimist(process.argv.slice(2)); const swaggerUrlFromCLI = argv['swagger-url'] || argv.swaggerUrl || null; // Store swagger URL in a way accessible to other modules if (swaggerUrlFromCLI) { let isValidUrl = false; try { // Throws if not a valid URL new URL(swaggerUrlFromCLI); isValidUrl = true; } catch (e) { logger.warn(`Invalid swagger-url provided via CLI: ${sanitizeForLog(swaggerUrlFromCLI)}`); } if (isValidUrl) { process.env.SWAGGER_URL_FROM_CLI = swaggerUrlFromCLI; logger.info(`Swagger URL from CLI: ${sanitizeForLog(swaggerUrlFromCLI)}`); } } // Import tool definitions and handlers import { toolDefinitions, handleGetSwaggerDefinition, handleListEndpoints, handleListEndpointModels, handleGenerateModelCode, handleGenerateEndpointToolCode, handleVersion } from "./tools/index.js"; // Import prompt definitions and handlers import { promptDefinitions, promptHandlers } from "./prompts/index.js"; // Create MCP server const server = new Server( { name: 'Swagger MCP Server', description: 'A server that helps you build a MCP wrapper around your Swagger API', version: '1.0.0', }, { capabilities: { tools: {}, prompts: {} }, } ); /** * Handler that lists available tools. * Exposes tools for interacting with Swagger API. */ server.setRequestHandler(ListToolsRequestSchema, async () => { return { tools: toolDefinitions }; }); /** * Handler that lists available prompts. * Exposes prompts for guiding through common workflows. */ server.setRequestHandler(ListPromptsRequestSchema, async () => { return { prompts: promptDefinitions }; }); /** * Handler for getting a specific prompt. * Returns the prompt template with messages. */ server.setRequestHandler(GetPromptRequestSchema, async (request) => { try { const promptName = request.params.name; const promptArgs = request.params.arguments || {}; logger.info(`Prompt request received: ${promptName}`); logger.info(`Prompt arguments: ${JSON.stringify(promptArgs)}`); const promptHandler = promptHandlers[promptName]; if (!promptHandler) { return { error: { code: -32601, message: `Unknown prompt: ${promptName}` } }; } // Validate arguments against schema const validationResult = promptHandler.schema.safeParse(promptArgs); if (!validationResult.success) { return { error: { code: -32602, message: `Invalid arguments: ${validationResult.error.message}` } }; } // Call the prompt handler const result = await promptHandler.handler(promptArgs); return result; } catch (error: any) { logger.error(`MCP prompt error: ${error.message}`); throw new Error(`Prompt execution failed: ${error.message}`); } }); /** * Handler for tool calls. * Processes requests to call Swagger API tools. */ server.setRequestHandler(CallToolRequestSchema, async (request) => { try { logger.info(`Tool call received: ${request.params.name}`); logger.info(`Tool arguments: ${JSON.stringify(request.params.arguments || {})}`); const name = request.params.name; const input = request.params.arguments; switch (name) { case "getSwaggerDefinition": return await handleGetSwaggerDefinition(input); case "listEndpoints": return await handleListEndpoints(input); case "listEndpointModels": return await handleListEndpointModels(input); case "generateModelCode": return await handleGenerateModelCode(input); case "generateEndpointToolCode": return await handleGenerateEndpointToolCode(input); case "version": return await handleVersion(input); default: return { content: [{ type: "text", text: `Unknown tool: ${name}` }] }; } } catch (error: any) { logger.error(`MCP tool error: ${error.message}`); throw new Error(`Tool execution failed: ${error.message}`); } }); /** * Start the server using stdio transport. * This allows the server to communicate via standard input/output streams. */ async function main() { // Connect using stdio transport const transport = new StdioServerTransport(); await server.connect(transport); } main().catch((error) => { logger.error("Server error:", error); process.exit(1); });