--- title: "Connect Sarvam to ChatGPT: Multilingual Speech and Text Processing" slug: connect-sarvam-to-chatgpt-multilingual-speech-and-text-processing date: 2026-06-19 author: Uday Gajavalli categories: ["AI & Agents"] excerpt: "Learn how to dynamically generate a managed MCP server for Sarvam, connect it to ChatGPT, and automate multilingual text, speech, and async job workflows." tldr: "Connect Sarvam to ChatGPT using Truto's generated MCP server to automate Indic language translations, speech-to-text batch processing, and TTS voice generation without building custom infrastructure." canonical: https://truto.one/blog/connect-sarvam-to-chatgpt-multilingual-speech-and-text-processing/ --- # Connect Sarvam to ChatGPT: Multilingual Speech and Text Processing If you need to connect Sarvam to ChatGPT to automate Indic language translation, speech-to-text transcription, or text-to-speech voice generation, you need a [Model Context Protocol (MCP) server](https://truto.one/what-is-mcp-and-mcp-servers-and-how-do-they-work/). This server acts as the translation layer between ChatGPT's tool calls and Sarvam's REST APIs. If your team uses Claude instead, check out our guide on [connecting Sarvam to Claude](https://truto.one/connect-sarvam-to-claude-indic-translation-and-audio-transcription/) or explore our broader architectural overview on [connecting Sarvam to AI Agents](https://truto.one/connect-sarvam-to-ai-agents-build-indic-voice-and-document-workflows/). Giving a Large Language Model (LLM) read and write access to a specialized regional AI platform like Sarvam is an engineering challenge. You have to handle multipart audio uploads, manage asynchronous job polling pipelines, map strict language code schemas, and deal with complex transliteration settings. Every time Sarvam updates an endpoint or changes their rate limits, you have to update your server code, redeploy, and test the integration. This guide breaks down exactly how to use Truto to generate a secure, managed MCP server for Sarvam, [connect it natively to ChatGPT](https://truto.one/bring-100-custom-connectors-to-chatgpt-with-superai-by-truto/), and execute complex multilingual workflows using natural language. ## The Engineering Reality of the Sarvam API A custom MCP server is a self-hosted integration layer. While the open MCP standard provides a predictable way for models to discover tools, the reality of implementing it against specialized AI models is painful. You aren't just doing simple database updates - you are dealing with binary files, asynchronous audio processing, and strict API rate limits. If you decide to build a custom MCP server for Sarvam, you own the entire API lifecycle. Here are the specific integration challenges that break standard CRUD assumptions when working with Sarvam: ### Asynchronous Batch STT Job Polling Sarvam's Saaras speech-to-text model uses an asynchronous batch processing architecture for large files. When an LLM wants to transcribe an hour of audio, it cannot just hold a connection open. It must initiate a job, receive a `job_id`, and continuously poll the `status` endpoint. The Sarvam documentation dictates a strict minimum delay (e.g., 5ms) between consecutive polling requests to avoid triggering abuse mechanisms. If your MCP server cannot instruct the LLM on exactly how to pace its queries, your agent will immediately burn through rate limits and fail. ### Audio File Uploads and Binary Responses The Sarvam API does not deal purely in JSON. Transcribing audio requires multipart form-data requests with the file payload intact. Conversely, generating audio via the Text-to-Speech (TTS) endpoint returns an opaque synthesized binary payload (or a stream). An LLM communicates primarily in text via JSON-RPC. Bridging this gap requires an integration layer that can properly handle and pass binary file references without corrupting the encoding or overflowing the LLM's context window. ### HTTP 429 Errors and Rate Limiting When processing intensive tasks like transliteration or translation across large text arrays, you will inevitably hit Sarvam's rate limits. It is critical to understand that Truto does not absorb, retry, or apply exponential backoff to these rate limit errors. When the upstream Sarvam API returns an HTTP 429, Truto passes that error directly to the caller. Truto does, however, normalize the upstream rate limit information into standardized headers (`ratelimit-limit`, `ratelimit-remaining`, `ratelimit-reset`) per the IETF specification. This means your MCP client (or ChatGPT agent) is strictly responsible for interpreting the `ratelimit-reset` header, pausing execution, and retrying the tool call. If your AI agent assumes the tool call succeeded without checking for 429 errors, it will hallucinate the translation. ```mermaid sequenceDiagram participant Agent as ChatGPT Agent participant MCP as Truto MCP Server participant Sarvam as Sarvam API Agent->>MCP: Call create_a_sarvam_text_translation MCP->>Sarvam: POST /translate Sarvam-->>MCP: 429 Too Many Requests MCP-->>Agent: Returns 429 + IETF ratelimit-reset headers Note over Agent: Agent must wait
and execute retry logic Agent->>MCP: Retry create_a_sarvam_text_translation MCP->>Sarvam: POST /translate Sarvam-->>MCP: 200 OK (Translated Text) MCP-->>Agent: JSON-RPC Success ``` ## Generating the Managed MCP Server Instead of building custom handlers for Sarvam's endpoints, Truto uses its unified proxy architecture to [dynamically generate MCP tools](https://truto.one/auto-generated-mcp-tools-for-ai-agents-a-2026-architecture-guide/). A tool only exists if it has a corresponding documentation and schema definition, acting as a strict quality gate. Tools are never pre-built - they are generated dynamically when the LLM requests them. There are two ways to generate an MCP server for Sarvam in Truto. ### Method 1: Via the Truto UI This is the fastest method for internal testing and one-off agent deployments. 1. Log into Truto and navigate to the **Integrated Accounts** page for your Sarvam connection. 2. Click the **MCP Servers** tab. 3. Click **Create MCP Server**. 4. Select your desired configuration (e.g., filtering for specific methods or tags). 5. Copy the generated secure MCP server URL. ### Method 2: Via the Truto API For production workflows where you need to programmatically provision AI agents for your customers, you can generate MCP servers via the API. The endpoint validates that your Sarvam connection is healthy, generates a cryptographic token hashed with a secure signing key, and stores it in an edge key-value store for ultra-low latency authentication. ```bash curl -X POST https://api.truto.one/integrated-account/{account_id}/mcp \ -H "Authorization: Bearer YOUR_TRUTO_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "name": "Sarvam Voice and Text MCP", "config": { "methods": ["read", "write", "custom"] } }' ``` The response contains the exact URL ChatGPT will use to authenticate and execute requests against the Sarvam API. ```json { "id": "mcp-srv-998877", "name": "Sarvam Voice and Text MCP", "url": "https://api.truto.one/mcp/a1b2c3d4e5f6..." } ``` ## Connecting the MCP Server to ChatGPT Once you have your Truto MCP URL, you must register it with ChatGPT so the LLM understands what Sarvam capabilities it has access to. ### Method 1: Via the ChatGPT UI If you are using ChatGPT Plus, Team, or Enterprise, you can connect the server natively through the application interface. 1. In ChatGPT, navigate to **Settings -> Apps -> Advanced settings**. 2. Toggle **Developer mode** to the "On" position. 3. Under the **MCP servers / Custom connectors** section, click Add a new server. 4. Name the connection (e.g., "Sarvam AI Server"). 5. Paste the Truto MCP URL you generated in the previous step. 6. Click **Save**. ChatGPT will immediately ping the server's `initialize` endpoint and retrieve the Sarvam tools via JSON-RPC. ### Method 2: Via Manual Configuration File If you are connecting via the Claude Desktop app or a custom development environment running an MCP host, you can configure the connection manually using the Server-Sent Events (SSE) transport protocol. Update your `mcp-config.json` file to point to the Truto URL: ```json { "mcpServers": { "sarvam_ai": { "command": "npx", "args": [ "-y", "@modelcontextprotocol/server-sse", "--url", "https://api.truto.one/mcp/a1b2c3d4e5f6..." ] } } } ``` ## Hero Tools for Sarvam When ChatGPT connects to Truto, it dynamically derives JSON-Schema definitions for Sarvam's endpoints. Here are the highest-leverage tools available for your AI agent. ### create_a_sarvam_text_translation Translates text from one Indic language to another. This is the core NLP tool for localizing conversational text across different regional dialects. > "I need to localize a support email. Take this English text: 'Your refund has been processed.' Call create_a_sarvam_text_translation to convert it to Hindi (`hi-IN`), then return the exact translated text." ### create_a_sarvam_speech_to_text Transcribes audio using Sarvam's Saaras v3 speech recognition model. This synchronous endpoint handles smaller files and supports multiple output modes, including transliteration and code-mixing. > "Take the attached customer voicemail file. Execute create_a_sarvam_speech_to_text. Ensure the output mode is set to 'translate' so I get an English transcription of the regional audio. Summarize the user's primary complaint from the transcript." ### create_a_sarvam_text_to_speech Converts a text string into spoken audio using Sarvam's TTS engine, synthesizing the output in the specified target language. Perfect for dynamic audio agent responses. > "Take the approved Hindi greeting script. Call create_a_sarvam_text_to_speech, passing the text and setting the target language code. Confirm when the binary payload is ready to be streamed back to the frontend." ### create_a_sarvam_speech_to_text_translate_job Initiates a batch asynchronous translation job for larger audio files. This tool returns a `job_id` instead of immediate text, requiring your agent to monitor the job status. > "We have a 45-minute webinar recording. Initiate create_a_sarvam_speech_to_text_translate_job with the file identifier. Note the `job_id` returned and pause execution before checking the status." ### get_single_sarvam_speech_to_text_translate_job_by_id Retrieves the status of an ongoing asynchronous translation job. Agents must be explicitly prompted to use this tool with careful polling intervals. > "Using the job_id from the previous step, call get_single_sarvam_speech_to_text_translate_job_by_id to check if the translation is complete. If the status is still processing, wait at least 10 seconds before polling again to respect the minimum delay rate limits." ### create_a_sarvam_text_language_identification Analyzes a raw text string and returns the identified language code. This is an essential routing tool for multi-language chatbots before they attempt translation or TTS. > "A user just submitted a ticket: 'ನನ್ನ ಆರ್ಡರ್ ಎಲ್ಲಿಯಿದೆ?'. Call create_a_sarvam_text_language_identification to determine the language code. Once identified, route it to the appropriate regional support queue." To view the complete inventory of tools, request parameters, and return schemas, visit the [Sarvam integration page](https://truto.one/integrations/detail/sarvam). ## Workflows in Action Giving ChatGPT access to these tools transforms it from a text generator into a multilingual command center. Here are real-world examples of how AI agents orchestrate Sarvam via Truto. ### Multilingual Customer Support Triage In an e-commerce support environment, customers frequently leave voice notes in mixed regional scripts. A human agent might take hours to translate and triage these. ChatGPT can handle it in seconds. > "A new customer voice note just arrived. First, call create_a_sarvam_speech_to_text using the 'codemix' output mode to transcribe the mixed-language audio. Next, call create_a_sarvam_text_language_identification on the transcript to find the dominant language. Finally, call create_a_sarvam_text_translation to convert the entire message to English. Summarize the issue and draft a suggested resolution." **Step-by-step execution:** 1. **create_a_sarvam_speech_to_text**: The agent uploads the file and receives a transcript that contains code-mixed vocabulary (e.g., English and Kannada). 2. **create_a_sarvam_text_language_identification**: The agent determines the base language is `kn-IN`. 3. **create_a_sarvam_text_translation**: The agent passes the code-mixed text and outputs clean English for the human support rep to review. ### Automated Webinar Localization Pipeline When a company hosts a 60-minute all-hands meeting, producing localized transcripts manually is expensive. An AI agent can manage the asynchronous Sarvam batch process autonomously. > "We need to localize the quarterly all-hands audio file into Tamil. Call create_a_sarvam_speech_to_text_translate_job to start the process. Use get_single_sarvam_speech_to_text_translate_job_by_id to check the status, ensuring you wait exactly 10 seconds between checks to respect rate limits. Once the transcript is returned, extract the top three key takeaways." **Step-by-step execution:** 1. **create_a_sarvam_speech_to_text_translate_job**: The agent posts the audio file and receives `job_id: "j-98765"`. 2. **get_single_sarvam_speech_to_text_translate_job_by_id**: The agent polls the endpoint. It may receive `status: "processing"`. 3. **get_single_sarvam_speech_to_text_translate_job_by_id (Retry)**: The agent waits, polls again, and receives `status: "completed"` with the full text payload. 4. **Local processing**: The LLM analyzes the final text and extracts the insights. ## Security and Access Control Exposing your Sarvam instance to an LLM requires strict governance. Truto's MCP architecture [enforces security through the URL token](https://truto.one/handling-auth-tool-sharing-in-multi-agent-frameworks-via-mcp/) itself, meaning the token defines exactly what the LLM is allowed to do. * **Method Filtering (`config.methods`)**: Restrict the AI to specific HTTP methods. For example, setting `methods: ["read"]` allows the agent to check job statuses and identify languages, but prevents it from creating new translation jobs that incur costs. * **Tag Filtering (`config.tags`)**: Scope tools by functional category. You can generate a server that only exposes tools tagged with `speech` while hiding all `document-intelligence` operations. * **Time-To-Live (`expires_at`)**: Generate ephemeral MCP servers for temporary workflows. By setting an `expires_at` timestamp, the edge token and underlying database record are automatically purged by distributed cleanup alarms, revoking access precisely when the window closes. * **Layered Authentication (`require_api_token_auth`)**: By default, possessing the MCP URL is enough to execute tools. For highly sensitive environments, enabling this flag forces the MCP client to also pass a valid Truto API token in the `Authorization` header, preventing unauthorized usage if the URL leaks. ## Moving Forward with Agentic Voice Workflows Building a custom MCP server to handle Sarvam's asynchronous batch polling and multipart audio uploads is a significant drain on engineering resources. Maintaining that connection when rate limits fluctuate or API schemas drift turns a one-time project into permanent technical debt. Truto eliminates this overhead by deriving tools dynamically from existing documentation schemas. Your engineers manage the integration connection, and the platform automatically provides a secure, properly scoped JSON-RPC interface for ChatGPT. The result is a system where your AI agents can seamlessly transcribe, translate, and synthesize multilingual audio without you having to write a single line of server code. > Stop wasting sprints building custom MCP servers for Sarvam. Use Truto's generated endpoints to connect AI agents to your specialized AI tools securely and at scale. > > [Talk to us](https://cal.com/truto/partner-with-truto)