@cencori/mcp 0.5.0 → 0.6.0

package/README.md

@@ -60,7 +60,7 @@ Docs + `how_to_*` guidance tools work with **no API key**. Add `CENCORI_API_KEY`
 
 ### Write — requires `CENCORI_MCP_WRITE=1`
 
-- **Inference:** `generate_text`, `generate_rag`, `create_embeddings`, `moderate_content`, `generate_image`, `describe_image`, `ocr_image`, `classify_image`, `extract_document`, `summarize_document`, `query_document`.
+- **Inference:** `generate_text`, `generate_rag`, `create_embeddings`, `moderate_content`, `generate_image`, `describe_image`, `ocr_image`, `classify_image`, `extract_document`, `summarize_document`, `query_document`, `text_to_speech`, `transcribe_audio`.
 - **Memory:** `remember_memory`, `write_memory`, `create_namespace`.
 - **Agents:** `create_agent`, `update_agent`.
 - **Sessions:** `create_session`, `add_session_turn`.

package/dist/index.js

@@ -148,6 +148,37 @@ var PlatformClient = class {
   del(path) {
     return this.request("DELETE", path);
   }
+  /** POST JSON and read a BINARY response (e.g. TTS audio). Returns base64 + mime. */
+  async postBinary(path, body) {
+    const url = new URL(`/api${path}`, this.baseUrl);
+    const response = await fetch(url, {
+      method: "POST",
+      headers: { Authorization: `Bearer ${this.apiKey}`, "Content-Type": "application/json" },
+      body: JSON.stringify(body),
+      signal: fetchSignal()
+    });
+    if (!response.ok) {
+      throw new Error(`Cencori API error: ${await readHttpErrorMessage(response)}`);
+    }
+    const mimeType = response.headers.get("content-type")?.split(";")[0]?.trim() || "application/octet-stream";
+    const buf = Buffer.from(await response.arrayBuffer());
+    return { base64: buf.toString("base64"), mimeType };
+  }
+  /** POST multipart/form-data (e.g. STT audio upload). Let fetch set the boundary. */
+  async postForm(path, form) {
+    const url = new URL(`/api${path}`, this.baseUrl);
+    const response = await fetch(url, {
+      method: "POST",
+      headers: { Authorization: `Bearer ${this.apiKey}` },
+      body: form,
+      signal: fetchSignal()
+    });
+    if (!response.ok) {
+      throw new Error(`Cencori API error: ${await readHttpErrorMessage(response)}`);
+    }
+    const text = await response.text();
+    return text ? JSON.parse(text) : void 0;
+  }
   listModels() {
     return this.get("/v1/models");
   }
@@ -1079,11 +1110,77 @@ function registerMultimodalTools(server, client) {
   );
 }
 
-// src/tools/guidance.ts
+// src/tools/audio.ts
 var import_zod8 = require("zod");
+function registerAudioTools(server, client) {
+  server.registerTool(
+    "text_to_speech",
+    {
+      title: "Text to speech",
+      description: "Synthesize speech from text. Returns audio. Incurs usage/cost.",
+      inputSchema: {
+        input: import_zod8.z.string().min(1).describe("The text to speak."),
+        model: import_zod8.z.string().optional().describe("TTS model. Defaults to tts-1."),
+        voice: import_zod8.z.string().optional().describe("Voice id/name (provider-specific)."),
+        response_format: import_zod8.z.string().optional().describe("Audio format, e.g. mp3, wav, opus."),
+        provider: import_zod8.z.string().optional().describe("Override the TTS provider.")
+      },
+      annotations: WRITE_ANNOTATIONS
+    },
+    async ({ input, model, voice, response_format, provider }) => {
+      const { base64, mimeType } = await client.postBinary("/ai/audio/speech", {
+        input,
+        model,
+        voice,
+        response_format,
+        provider
+      });
+      return {
+        content: [
+          { type: "audio", data: base64, mimeType },
+          { type: "text", text: `Synthesized ${input.length} chars \u2192 ${mimeType} (${base64.length} b64 bytes).` }
+        ]
+      };
+    }
+  );
+  server.registerTool(
+    "transcribe_audio",
+    {
+      title: "Transcribe audio (speech to text)",
+      description: "Transcribe base64-encoded audio to text. Incurs usage/cost.",
+      inputSchema: {
+        audio_base64: import_zod8.z.string().min(1).describe("Base64-encoded audio bytes."),
+        mime_type: import_zod8.z.string().optional().describe("Audio MIME type, e.g. audio/mpeg, audio/wav."),
+        filename: import_zod8.z.string().optional().describe("Original filename (helps format detection)."),
+        model: import_zod8.z.string().optional().describe("STT model. Defaults to whisper-1."),
+        language: import_zod8.z.string().optional().describe("ISO language code, e.g. en."),
+        prompt: import_zod8.z.string().optional().describe("Optional prompt to guide transcription."),
+        response_format: import_zod8.z.enum(["json", "text", "srt", "verbose_json", "vtt"]).optional().describe("Transcript format. verbose_json adds segments/words."),
+        temperature: import_zod8.z.number().min(0).max(1).optional(),
+        provider: import_zod8.z.string().optional().describe("Override the STT provider.")
+      },
+      annotations: WRITE_ANNOTATIONS
+    },
+    async ({ audio_base64, mime_type, filename, model, language, prompt, response_format, temperature, provider }) => {
+      const bytes = Buffer.from(audio_base64, "base64");
+      const form = new FormData();
+      form.append("file", new Blob([bytes], { type: mime_type || "application/octet-stream" }), filename || "audio");
+      if (model) form.append("model", model);
+      if (language) form.append("language", language);
+      if (prompt) form.append("prompt", prompt);
+      if (response_format) form.append("response_format", response_format);
+      if (temperature !== void 0) form.append("temperature", String(temperature));
+      if (provider) form.append("provider", provider);
+      return jsonResult(await client.postForm("/ai/audio/transcriptions", form));
+    }
+  );
+}
+
+// src/tools/guidance.ts
+var import_zod9 = require("zod");
 var orgProjectShape = {
-  org_slug: import_zod8.z.string().min(1).optional().describe("Your organization slug, to build an exact dashboard link."),
-  project_slug: import_zod8.z.string().min(1).optional().describe("Your project slug, to build an exact dashboard link.")
+  org_slug: import_zod9.z.string().min(1).optional().describe("Your organization slug, to build an exact dashboard link."),
+  project_slug: import_zod9.z.string().min(1).optional().describe("Your project slug, to build an exact dashboard link.")
 };
 function registerGuide(server, name, title, description, buildPath, steps, baseUrl) {
   server.registerTool(
@@ -1214,7 +1311,7 @@ function registerGuidanceTools(server, baseUrl) {
 
 // src/server.ts
 var SERVER_NAME = "cencori";
-var SERVER_VERSION = "0.5.0";
+var SERVER_VERSION = "0.6.0";
 function createServer(config) {
   const server = new import_mcp.McpServer(
     {
@@ -1245,6 +1342,7 @@ function createServer(config) {
     if (features.governance) registerGovernanceTools(server, client, capabilities);
     if (features.multimodal && capabilities.write) {
       registerMultimodalTools(server, client);
+      registerAudioTools(server, client);
     }
   }
   return server;