@hotmeshio/long-tail 0.1.4 → 0.1.6

package/build/system/mcp-servers/playwright-cli/tools-capture.js

@@ -35,7 +35,7 @@ var __importStar = (this && this.__importStar) || (function () {
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.registerCaptureTools = registerCaptureTools;
 const path = __importStar(require("path"));
-const logger_1 = require("../../../services/logger");
+const logger_1 = require("../../../lib/logger");
 const playwright_1 = require("../playwright");
 const schemas_1 = require("./schemas");
 const helpers_1 = require("./helpers");
@@ -43,7 +43,8 @@ function registerCaptureTools(srv) {
     // capture_page
     srv.registerTool('capture_page', {
         title: 'Capture Page',
-        description: 'Navigate to a URL and capture a screenshot in one call. ' +
+        description: 'Navigate to a URL, capture a screenshot, and save it to persistent file storage in one call. ' +
+            'The returned `path` is the storage reference — no separate write_file call is needed. ' +
             'Optionally waits for a CSS selector or a fixed delay to ensure content is loaded. ' +
             'Pass page_id from a prior login_and_capture to reuse the authenticated session.',
         inputSchema: schemas_1.capturePageSchema,
@@ -83,7 +84,7 @@ function registerCaptureTools(srv) {
                 await page.waitForTimeout(args.wait_ms ?? 2000);
             }
             // Auto-derive screenshot path from URL if not provided
-            const screenshotPath = args.screenshot_path || (() => {
+            let screenshotPath = args.screenshot_path || (() => {
                 try {
                     const urlPath = new URL(args.url).pathname;
                     const slug = urlPath === '/' ? 'home' : urlPath.replace(/^\/+|\/+$/g, '').replace(/\//g, '-');
@@ -93,6 +94,10 @@ function registerCaptureTools(srv) {
                     return `screenshots/capture-${Date.now()}.png`;
                 }
             })();
+            // Ensure file extension — storage backends require mime type
+            if (!path.extname(screenshotPath)) {
+                screenshotPath += '.png';
+            }
             const screenshot = await (0, helpers_1.saveScreenshot)(page, screenshotPath, args.full_page ?? true);
             return {
                 content: [{
@@ -102,6 +107,7 @@ function registerCaptureTools(srv) {
                             url: page.url(),
                             title: await page.title(),
                             ...screenshot,
+                            storage_ref: screenshot.path,
                             _handle: (0, playwright_1.buildHandle)(pageId),
                         }),
                     }],

package/build/system/mcp-servers/translation.d.ts

@@ -0,0 +1,14 @@
+import { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js';
+/**
+ * Create the Translation MCP server.
+ *
+ * Returns a fresh McpServer instance each time. The MCP SDK only allows
+ * one transport per server, so each consumer needs its own instance.
+ */
+export declare function createTranslationServer(options?: {
+    name?: string;
+}): Promise<McpServer>;
+/**
+ * Stop a Translation MCP server instance.
+ */
+export declare function stopTranslationServer(): Promise<void>;

package/build/system/mcp-servers/translation.js

@@ -0,0 +1,130 @@
+"use strict";
+var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    var desc = Object.getOwnPropertyDescriptor(m, k);
+    if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
+      desc = { enumerable: true, get: function() { return m[k]; } };
+    }
+    Object.defineProperty(o, k2, desc);
+}) : (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    o[k2] = m[k];
+}));
+var __setModuleDefault = (this && this.__setModuleDefault) || (Object.create ? (function(o, v) {
+    Object.defineProperty(o, "default", { enumerable: true, value: v });
+}) : function(o, v) {
+    o["default"] = v;
+});
+var __importStar = (this && this.__importStar) || (function () {
+    var ownKeys = function(o) {
+        ownKeys = Object.getOwnPropertyNames || function (o) {
+            var ar = [];
+            for (var k in o) if (Object.prototype.hasOwnProperty.call(o, k)) ar[ar.length] = k;
+            return ar;
+        };
+        return ownKeys(o);
+    };
+    return function (mod) {
+        if (mod && mod.__esModule) return mod;
+        var result = {};
+        if (mod != null) for (var k = ownKeys(mod), i = 0; i < k.length; i++) if (k[i] !== "default") __createBinding(result, mod, k[i]);
+        __setModuleDefault(result, mod);
+        return result;
+    };
+})();
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.createTranslationServer = createTranslationServer;
+exports.stopTranslationServer = stopTranslationServer;
+const mcp_js_1 = require("@modelcontextprotocol/sdk/server/mcp.js");
+const zod_1 = require("zod");
+const llm_1 = require("../../services/llm");
+const defaults_1 = require("../../modules/defaults");
+const logger_1 = require("../../lib/logger");
+function TRANSLATE_SYSTEM_PROMPT(targetLanguage) {
+    return `You are a translation assistant. Translate the user's text to ${targetLanguage}. Return ONLY a JSON object: {"translated_content": "...", "source_language": "detected ISO code"}. No markdown, no explanation.`;
+}
+const translateContentSchema = zod_1.z.object({
+    content: zod_1.z.string().describe('The content text to translate'),
+    target_language: zod_1.z.string().describe('Target language code (e.g. "en", "es")'),
+    source_language: zod_1.z.string().optional().describe('Source language code (auto-detected if omitted)'),
+});
+function registerTools(srv) {
+    srv.registerTool('translate_content', {
+        title: 'Translate Content',
+        description: 'Translate content text to the target language. Returns the translated content and detected source language.',
+        inputSchema: translateContentSchema,
+    }, async (args) => {
+        const { hasLLMApiKey } = await Promise.resolve().then(() => __importStar(require('../../services/llm')));
+        if (!hasLLMApiKey(defaults_1.LLM_MODEL_SECONDARY)) {
+            return {
+                content: [{
+                        type: 'text',
+                        text: JSON.stringify({
+                            translated_content: args.content,
+                            source_language: args.source_language || 'unknown',
+                            target_language: args.target_language,
+                            note: 'LLM API key not configured — returned content unchanged',
+                        }),
+                    }],
+            };
+        }
+        const systemPrompt = args.source_language
+            ? `You are a translation assistant. Translate the user's text from ${args.source_language} to ${args.target_language}. Return ONLY a JSON object: {"translated_content": "...", "source_language": "${args.source_language}"}. No markdown, no explanation.`
+            : TRANSLATE_SYSTEM_PROMPT(args.target_language);
+        const response = await (0, llm_1.callLLM)({
+            model: defaults_1.LLM_MODEL_SECONDARY,
+            messages: [
+                { role: 'system', content: systemPrompt },
+                { role: 'user', content: args.content },
+            ],
+            max_tokens: defaults_1.LLM_MAX_TOKENS_VISION,
+            temperature: 0,
+        });
+        const raw = response.content || '';
+        try {
+            const cleaned = raw.replace(/^```json\n?|\n?```$/g, '').trim();
+            const parsed = JSON.parse(cleaned);
+            return {
+                content: [{
+                        type: 'text',
+                        text: JSON.stringify({
+                            translated_content: parsed.translated_content || args.content,
+                            source_language: parsed.source_language || args.source_language || 'unknown',
+                            target_language: args.target_language,
+                        }),
+                    }],
+            };
+        }
+        catch {
+            return {
+                content: [{
+                        type: 'text',
+                        text: JSON.stringify({
+                            translated_content: raw,
+                            source_language: args.source_language || 'unknown',
+                            target_language: args.target_language,
+                        }),
+                    }],
+            };
+        }
+    });
+}
+/**
+ * Create the Translation MCP server.
+ *
+ * Returns a fresh McpServer instance each time. The MCP SDK only allows
+ * one transport per server, so each consumer needs its own instance.
+ */
+async function createTranslationServer(options) {
+    const name = options?.name || 'long-tail-translation';
+    const instance = new mcp_js_1.McpServer({ name, version: '1.0.0' });
+    registerTools(instance);
+    logger_1.loggerRegistry.info(`[lt-mcp:translation] ${name} ready (1 tool registered)`);
+    return instance;
+}
+/**
+ * Stop a Translation MCP server instance.
+ */
+async function stopTranslationServer() {
+    logger_1.loggerRegistry.info('[lt-mcp:translation] stopped');
+}

package/build/system/mcp-servers/vision-prompts.d.ts

@@ -0,0 +1,2 @@
+export declare const ANALYZE_IMAGE_PROMPT = "You are an image analysis assistant. Analyze the provided image and return a JSON object with the following structure:\n{\"description\": \"brief description of the image\", \"text_content\": \"any text visible in the image or null\", \"objects\": [\"list\", \"of\", \"notable\", \"objects\"]}\nReturn ONLY the JSON object. No markdown, no explanation.";
+export declare const DESCRIBE_IMAGE_PROMPT = "You are an image description assistant. Provide a clear, detailed description of the provided image. Return a JSON object:\n{\"description\": \"detailed description of the image\"}\nReturn ONLY the JSON object. No markdown, no explanation.";

package/build/system/mcp-servers/vision-prompts.js

@@ -0,0 +1,9 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.DESCRIBE_IMAGE_PROMPT = exports.ANALYZE_IMAGE_PROMPT = void 0;
+exports.ANALYZE_IMAGE_PROMPT = `You are an image analysis assistant. Analyze the provided image and return a JSON object with the following structure:
+{"description": "brief description of the image", "text_content": "any text visible in the image or null", "objects": ["list", "of", "notable", "objects"]}
+Return ONLY the JSON object. No markdown, no explanation.`;
+exports.DESCRIBE_IMAGE_PROMPT = `You are an image description assistant. Provide a clear, detailed description of the provided image. Return a JSON object:
+{"description": "detailed description of the image"}
+Return ONLY the JSON object. No markdown, no explanation.`;

package/build/system/mcp-servers/{document-vision.d.ts → vision.d.ts}

@@ -1,15 +1,14 @@
 import { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js';
 /**
- * Create the Document Vision MCP server.
+ * Create the Vision MCP server.
  *
  * Returns a fresh McpServer instance each time. The MCP SDK only allows
- * one transport per server, so each consumer (triage activities, pipeline
- * workers, tests) needs its own instance to avoid "already connected" errors.
+ * one transport per server, so each consumer needs its own instance.
  */
 export declare function createVisionServer(options?: {
     name?: string;
 }): Promise<McpServer>;
 /**
- * Stop a Vision MCP server instance and release resources.
+ * Stop a Vision MCP server instance.
  */
 export declare function stopVisionServer(): Promise<void>;

package/build/system/mcp-servers/vision.js

@@ -0,0 +1,258 @@
+"use strict";
+var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    var desc = Object.getOwnPropertyDescriptor(m, k);
+    if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
+      desc = { enumerable: true, get: function() { return m[k]; } };
+    }
+    Object.defineProperty(o, k2, desc);
+}) : (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    o[k2] = m[k];
+}));
+var __setModuleDefault = (this && this.__setModuleDefault) || (Object.create ? (function(o, v) {
+    Object.defineProperty(o, "default", { enumerable: true, value: v });
+}) : function(o, v) {
+    o["default"] = v;
+});
+var __importStar = (this && this.__importStar) || (function () {
+    var ownKeys = function(o) {
+        ownKeys = Object.getOwnPropertyNames || function (o) {
+            var ar = [];
+            for (var k in o) if (Object.prototype.hasOwnProperty.call(o, k)) ar[ar.length] = k;
+            return ar;
+        };
+        return ownKeys(o);
+    };
+    return function (mod) {
+        if (mod && mod.__esModule) return mod;
+        var result = {};
+        if (mod != null) for (var k = ownKeys(mod), i = 0; i < k.length; i++) if (k[i] !== "default") __createBinding(result, mod, k[i]);
+        __setModuleDefault(result, mod);
+        return result;
+    };
+})();
+var __importDefault = (this && this.__importDefault) || function (mod) {
+    return (mod && mod.__esModule) ? mod : { "default": mod };
+};
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.createVisionServer = createVisionServer;
+exports.stopVisionServer = stopVisionServer;
+const path = __importStar(require("path"));
+const mcp_js_1 = require("@modelcontextprotocol/sdk/server/mcp.js");
+const zod_1 = require("zod");
+const sharp_1 = __importDefault(require("sharp"));
+const llm_1 = require("../../services/llm");
+const defaults_1 = require("../../modules/defaults");
+const logger_1 = require("../../lib/logger");
+const storage_1 = require("../../lib/storage");
+const vision_prompts_1 = require("./vision-prompts");
+const MIME_MAP = {
+    '.png': 'image/png', '.jpg': 'image/jpeg', '.jpeg': 'image/jpeg',
+    '.gif': 'image/gif', '.webp': 'image/webp', '.svg': 'image/svg+xml',
+};
+/** Max dimension for LLM vision APIs (Anthropic limit is 8000px). */
+const MAX_IMAGE_DIMENSION = 7680;
+const analyzeImageSchema = zod_1.z.object({
+    image: zod_1.z.string().describe('Image source: a storage path (e.g. "amazon_initial.png"), ' +
+        'a data URI (data:image/png;base64,...), or an https:// URL.'),
+    prompt: zod_1.z.string().optional().describe('Optional analysis prompt to guide the model'),
+});
+const describeImageSchema = zod_1.z.object({
+    image: zod_1.z.string().describe('Image source: a storage path, data URI, or https:// URL.'),
+    context: zod_1.z.string().optional().describe('Optional context about the image'),
+});
+/**
+ * Resolve an image reference to an LLM-ready content block.
+ *
+ * Accepts:
+ * - data:image/...;base64,... → used directly
+ * - https:// URL → passed as-is (OpenAI can fetch; Anthropic translate layer handles)
+ * - Storage path (anything else, including file:// prefixed) → read via storage backend
+ */
+async function resolveImageContent(image) {
+    // Already a data URI — pass through
+    if (image.startsWith('data:')) {
+        return { type: 'image_url', image_url: { url: image } };
+    }
+    // HTTPS URL — pass through
+    if (image.startsWith('https://') || image.startsWith('http://')) {
+        return { type: 'image_url', image_url: { url: image } };
+    }
+    // Storage path — strip file:// prefix if present, read via backend
+    const storagePath = image.replace(/^file:\/\//, '');
+    const backend = (0, storage_1.getStorageBackend)();
+    const { data } = await backend.read(storagePath);
+    // Downscale if any dimension exceeds the LLM vision limit
+    let imageBuffer = data;
+    let outputMime = MIME_MAP[path.extname(storagePath).toLowerCase()] || 'image/png';
+    try {
+        const meta = await (0, sharp_1.default)(data).metadata();
+        if (meta.width && meta.height && (meta.width > MAX_IMAGE_DIMENSION || meta.height > MAX_IMAGE_DIMENSION)) {
+            logger_1.loggerRegistry.info(`[lt-mcp:vision] resizing ${meta.width}x${meta.height} -> fit ${MAX_IMAGE_DIMENSION}px`);
+            imageBuffer = await (0, sharp_1.default)(data)
+                .resize({ width: MAX_IMAGE_DIMENSION, height: MAX_IMAGE_DIMENSION, fit: 'inside' })
+                .png()
+                .toBuffer();
+            outputMime = 'image/png';
+        }
+    }
+    catch (err) {
+        logger_1.loggerRegistry.warn(`[lt-mcp:vision] could not check/resize image: ${err.message}`);
+    }
+    const base64 = imageBuffer.toString('base64');
+    return { type: 'image_url', image_url: { url: `data:${outputMime};base64,${base64}` } };
+}
+function registerTools(srv) {
+    // ── analyze_image ───────────────────────────────────────────────
+    srv.registerTool('analyze_image', {
+        title: 'Analyze Image',
+        description: 'Analyze an image and extract structured data: description, text content, and notable objects.',
+        inputSchema: analyzeImageSchema,
+    }, async (args) => {
+        const { hasLLMApiKey } = await Promise.resolve().then(() => __importStar(require('../../services/llm')));
+        if (!hasLLMApiKey(defaults_1.LLM_MODEL_SECONDARY)) {
+            return {
+                content: [{
+                        type: 'text',
+                        text: JSON.stringify({
+                            error: 'LLM API key not configured',
+                            description: null,
+                            text_content: null,
+                            objects: [],
+                        }),
+                    }],
+            };
+        }
+        const systemPrompt = args.prompt
+            ? `${vision_prompts_1.ANALYZE_IMAGE_PROMPT}\n\nAdditional guidance: ${args.prompt}`
+            : vision_prompts_1.ANALYZE_IMAGE_PROMPT;
+        let response;
+        try {
+            const imageContent = await resolveImageContent(args.image || args.image_path);
+            response = await (0, llm_1.callLLM)({
+                model: defaults_1.LLM_MODEL_SECONDARY,
+                messages: [
+                    { role: 'system', content: systemPrompt },
+                    { role: 'user', content: [imageContent] },
+                ],
+                max_tokens: defaults_1.LLM_MAX_TOKENS_VISION,
+                temperature: 0,
+            });
+        }
+        catch (err) {
+            logger_1.loggerRegistry.error(`[lt-mcp:vision] analyze_image failed: ${err.message}`);
+            return {
+                content: [{
+                        type: 'text',
+                        text: JSON.stringify({
+                            error: err.message,
+                            description: null,
+                            text_content: null,
+                            objects: [],
+                        }),
+                    }],
+            };
+        }
+        const raw = response.content || '';
+        logger_1.loggerRegistry.info(`[lt-mcp:vision] analyze_image response length: ${raw.length}`);
+        try {
+            const cleaned = raw.replace(/^```json\n?|\n?```$/g, '').trim();
+            const parsed = JSON.parse(cleaned);
+            return {
+                content: [{
+                        type: 'text',
+                        text: JSON.stringify({
+                            description: parsed.description || null,
+                            text_content: parsed.text_content || null,
+                            objects: parsed.objects || [],
+                        }),
+                    }],
+            };
+        }
+        catch {
+            return {
+                content: [{
+                        type: 'text',
+                        text: JSON.stringify({
+                            description: raw,
+                            text_content: null,
+                            objects: [],
+                        }),
+                    }],
+            };
+        }
+    });
+    // ── describe_image ──────────────────────────────────────────────
+    srv.registerTool('describe_image', {
+        title: 'Describe Image',
+        description: 'Generate a detailed description of an image.',
+        inputSchema: describeImageSchema,
+    }, async (args) => {
+        const { hasLLMApiKey } = await Promise.resolve().then(() => __importStar(require('../../services/llm')));
+        if (!hasLLMApiKey(defaults_1.LLM_MODEL_SECONDARY)) {
+            return {
+                content: [{
+                        type: 'text',
+                        text: JSON.stringify({
+                            error: 'LLM API key not configured',
+                            description: null,
+                        }),
+                    }],
+            };
+        }
+        const systemPrompt = args.context
+            ? `${vision_prompts_1.DESCRIBE_IMAGE_PROMPT}\n\nContext: ${args.context}`
+            : vision_prompts_1.DESCRIBE_IMAGE_PROMPT;
+        const response = await (0, llm_1.callLLM)({
+            model: defaults_1.LLM_MODEL_SECONDARY,
+            messages: [
+                { role: 'system', content: systemPrompt },
+                {
+                    role: 'user',
+                    content: [await resolveImageContent(args.image || args.image_path)],
+                },
+            ],
+            max_tokens: defaults_1.LLM_MAX_TOKENS_VISION,
+            temperature: 0,
+        });
+        const raw = response.content || '';
+        try {
+            const cleaned = raw.replace(/^```json\n?|\n?```$/g, '').trim();
+            const parsed = JSON.parse(cleaned);
+            return {
+                content: [{
+                        type: 'text',
+                        text: JSON.stringify({ description: parsed.description || raw }),
+                    }],
+            };
+        }
+        catch {
+            return {
+                content: [{
+                        type: 'text',
+                        text: JSON.stringify({ description: raw }),
+                    }],
+            };
+        }
+    });
+}
+/**
+ * Create the Vision MCP server.
+ *
+ * Returns a fresh McpServer instance each time. The MCP SDK only allows
+ * one transport per server, so each consumer needs its own instance.
+ */
+async function createVisionServer(options) {
+    const name = options?.name || 'long-tail-vision';
+    const instance = new mcp_js_1.McpServer({ name, version: '1.0.0' });
+    registerTools(instance);
+    logger_1.loggerRegistry.info(`[lt-mcp:vision] ${name} ready (2 tools registered)`);
+    return instance;
+}
+/**
+ * Stop a Vision MCP server instance.
+ */
+async function stopVisionServer() {
+    logger_1.loggerRegistry.info('[lt-mcp:vision] stopped');
+}

package/build/system/mcp-servers/workflow-compiler.js

@@ -38,7 +38,7 @@ exports.getWorkflowCompilerServer = getWorkflowCompilerServer;
 exports.stopWorkflowCompilerServer = stopWorkflowCompilerServer;
 const mcp_js_1 = require("@modelcontextprotocol/sdk/server/mcp.js");
 const zod_1 = require("zod");
-const logger_1 = require("../../services/logger");
+const logger_1 = require("../../lib/logger");
 const yamlGenerator = __importStar(require("../../services/yaml-workflow/generator"));
 const yamlDb = __importStar(require("../../services/yaml-workflow/db"));
 const yamlDeployer = __importStar(require("../../services/yaml-workflow/deployer"));

package/build/system/mcp-servers/workflow.js

@@ -39,7 +39,7 @@ exports.stopWorkflowServer = stopWorkflowServer;
 const mcp_js_1 = require("@modelcontextprotocol/sdk/server/mcp.js");
 const zod_1 = require("zod");
 const defaults_1 = require("../../modules/defaults");
-const logger_1 = require("../../services/logger");
+const logger_1 = require("../../lib/logger");
 const yamlDb = __importStar(require("../../services/yaml-workflow/db"));
 const yamlDeployer = __importStar(require("../../services/yaml-workflow/deployer"));
 let server = null;

package/build/system/seed/index.js

@@ -2,8 +2,8 @@
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.SEED_MCP_SERVERS = void 0;
 exports.seedSystemMcpServers = seedSystemMcpServers;
-const db_1 = require("../../services/db");
-const logger_1 = require("../../services/logger");
+const db_1 = require("../../lib/db");
+const logger_1 = require("../../lib/logger");
 const sql_1 = require("../../services/mcp/sql");
 const server_definitions_1 = require("./server-definitions");
 Object.defineProperty(exports, "SEED_MCP_SERVERS", { enumerable: true, get: function () { return server_definitions_1.SEED_MCP_SERVERS; } });
@@ -31,6 +31,17 @@ async function seedSystemMcpServers() {
             logger_1.loggerRegistry.warn(`[system] failed to seed MCP server ${srv.name}: ${err.message}`);
         }
     }
+    // Remove stale builtin servers no longer in the seed list
+    const seedNames = server_definitions_1.SEED_MCP_SERVERS.map(s => s.name);
+    try {
+        const { rows } = await pool.query(sql_1.DELETE_STALE_BUILTIN_SERVERS, [seedNames]);
+        for (const row of rows) {
+            logger_1.loggerRegistry.info(`[system] removed stale builtin MCP server: ${row.name}`);
+        }
+    }
+    catch (err) {
+        logger_1.loggerRegistry.warn(`[system] failed to clean stale MCP servers: ${err.message}`);
+    }
     const totalTools = server_definitions_1.SEED_MCP_SERVERS.reduce((sum, s) => sum + s.tool_manifest.length, 0);
     logger_1.loggerRegistry.info(`[system] MCP servers seeded (${server_definitions_1.SEED_MCP_SERVERS.length} servers, ${totalTools} tools)`);
 }