Add NexusAI: universal MCP service (TTS/STT/image/video/music/web-search)

Monorepo-lite with shared @nexusai/core and @nexusai/mcp-server (stdio). Provider-agnostic tools with explicit provider selection; NordRouter adapter (media generate/poll/download, estimate, upload, models) and search via perplexity/sonar. Local STT via faster-whisper uv sidecar. SQLite journal of every generation for usage stats and future admin. 10 MCP tools.
2026-08-17 17:06:47 +07:00
parent 82087be1b3
commit 02fc536e85
32 changed files with 3861 additions and 0 deletions
@@ -0,0 +1,78 @@
+import { zodToJsonSchema } from 'zod-to-json-schema';
+import {
+  ImageSchema,
+  MediaEstimateSchema,
+  MediaModelsSchema,
+  MediaUploadSchema,
+  MusicSchema,
+  SttSchema,
+  TtsSchema,
+  UsageStatsSchema,
+  VideoSchema,
+  WebSearchSchema,
+} from './schemas.js';
+
+export interface ToolDef {
+  name: string;
+  description: string;
+  inputSchema: object;
+}
+
+function make(name: string, description: string, schema: unknown): ToolDef {
+  const json = zodToJsonSchema(schema as any, { name, $refStrategy: 'none' }) as any;
+  const def = json.definitions?.[name] ?? json;
+  return { name, description, inputSchema: def };
+}
+
+export const TOOLS: ToolDef[] = [
+  make(
+    'nexus_tts',
+    'Text-to-speech. Generates spoken audio from text via a provider. Returns a saved audio file path and optional embedded audio.',
+    TtsSchema
+  ),
+  make(
+    'nexus_image',
+    'Generate or edit images from a text prompt (and optional source image URL). Returns a saved image file and optional embedded image.',
+    ImageSchema
+  ),
+  make(
+    'nexus_video',
+    'Generate video from text or an image (t2v / i2v). Async, may take minutes. Returns a saved MP4 file path.',
+    VideoSchema
+  ),
+  make(
+    'nexus_music',
+    'Generate music from a style/prompt. Returns saved audio file(s); some models produce two tracks.',
+    MusicSchema
+  ),
+  make(
+    'nexus_stt',
+    'Speech-to-text. Transcribe local/remote/base64 audio using a local whisper model. Returns text, language and segments.',
+    SttSchema
+  ),
+  make(
+    'nexus_web_search',
+    'Web search that returns an answer with cited source URLs (via search-augmented models).',
+    WebSearchSchema
+  ),
+  make(
+    'nexus_media_models',
+    'List available media models (id, type, label, estimated price) for a provider. Use to discover valid model ids and parameters.',
+    MediaModelsSchema
+  ),
+  make(
+    'nexus_media_estimate',
+    'Estimate the USD cost of a media generation before running it.',
+    MediaEstimateSchema
+  ),
+  make(
+    'nexus_media_upload',
+    'Upload a local media file to the provider and get a reference URL for use as input (image/audio/video).',
+    MediaUploadSchema
+  ),
+  make(
+    'nexus_usage_stats',
+    'Show usage and spend recorded in the local journal (totals by provider, capability, day) and optionally recent generations.',
+    UsageStatsSchema
+  ),
+];