Groq MCP Server: Models, Audio and Batch Jobs

💡 No install? Use cloud.anythingmcp.com directly. Sign in, click Connectors → Groq, paste your credentials, mint an MCP API key and you're done. No Docker, no git clone, no local server.

Groq

GroqCloud with your own API key: fast chat completions from open models such as Llama, GPT-OSS and Qwen, speech-to-text and translation into English of audio files from a public link with Whisper, and the models, batch jobs and files of the organization.

Setup

Setting up takes about 2 minutes. You need a GroqCloud account; the free plan works within its rate limits.

1. Sign in at https://console.groq.com and open API Keys (https://console.groq.com/keys).
2. Click Create API Key, give it a name and copy it. It starts with gsk_ and Groq shows it only once.
3. Paste it into GROQ_API_KEY, without the word Bearer. The connector sends it as Authorization: Bearer and only to Groq.
4. Install the connector and run groq_list_models: it lists the models the key can use and confirms the key.

Which tools cost money: groq_chat_completion, groq_transcribe_audio and groq_translate_audio run a model. On the free plan they are free but rate limited; on the Developer plan every call is billed to the organization: text per input and output token, Whisper per hour of audio with a minimum of 10 seconds per request. The AI asks you before large or repeated calls and sets max_completion_tokens to cap a reply. groq_list_models, groq_get_model, groq_list_batches, groq_get_batch, groq_list_files and groq_get_file cost nothing.

Models: groq_list_models shows every model the key can use with owner, context_window, max_completion_tokens and whether it is active; groq_get_model reads one and answers 404 for a model the key cannot use. Model ids change often: preview models can be removed at short notice, production models are stable. Common ids are llama-3.3-70b-versatile, llama-3.1-8b-instant (the cheapest), openai/gpt-oss-120b, openai/gpt-oss-20b and qwen/qwen3.8-27b; for audio whisper-large-v3-turbo (cheaper) and whisper-large-v3.

Text: groq_chat_completion takes model and messages and returns the whole reply at once (no streaming). response_format asks for JSON or a JSON schema, tools adds function calling, and reasoning_effort and reasoning_format steer reasoning models (GPT-OSS takes low, medium or high; Qwen takes none or default).

Audio: groq_transcribe_audio writes speech down in its own language, groq_translate_audio turns it into English text (only whisper-large-v3 translates). Both take a public link in url: Groq downloads the file itself, so nothing is uploaded. Formats: flac, mp3, mp4, mpeg, mpga, m4a, ogg, wav, webm. Passing language (ISO-639-1, for example de) makes it faster and more accurate. response_format verbose_json adds segments with timestamps, timestamp_granularities word adds word timings.

Batch jobs and files: groq_list_batches and groq_get_batch show batches started elsewhere with their status (validating, in_progress, finalizing, completed, failed, expired, cancelling, cancelled), request counts and the output_file_id and error_file_id. groq_list_files and groq_get_file show file name, size in bytes, purpose (batch or batch_output) and creation time, not the content.

Data formats: a completion returns choices[0].message.content, finish_reason (stop, length when max_completion_tokens was reached, tool_calls) and usage with token counts and timings in seconds; reasoning models may put their reasoning in choices[0].message.reasoning. The result also carries _headers with the requests and tokens left in the rate limit. A transcription returns text. Timestamps such as created are Unix seconds, UTC.

Paging: the list tools have no paging parameters; they return what Groq sends in one answer.

Switched off at install: nothing. All 9 tools are on, and none of them deletes anything.

Limits: rate limits apply per organization and model, as requests and tokens per minute and per day (https://console.groq.com/settings/limits). Audio links may point to files of up to 25 MB on the free plan and 100 MB on paid plans.

Common errors: 400 means a parameter the model does not support or audio that cannot be read; 401 a wrong or deleted key; 403 the project may not use this model; 404 an unknown model, batch or file; 413 an audio file that is too large; 429 the rate limit is reached (the AI waits the seconds in retry-after instead of retrying in a loop); 498 the flex tier is out of capacity; 500 and 503 are Groq problems, retry once later.

The connector does not stream replies, does not upload or download file content, and does not create or cancel batch jobs.

Prerequisites: GROQ_API_KEY

Install the connector locally

curl -fsSL https://raw.githubusercontent.com/HelpCode-ai/anythingmcp/main/docker-compose.quickstart.yml -o docker-compose.yml
printf 'JWT_SECRET=%s\nENCRYPTION_KEY=%s\n' "$(openssl rand -hex 32)" "$(openssl rand -hex 32)" > .env
docker compose up -d

Open http://localhost:3000/connectors/store, pick Groq and paste in the env vars listed under Prerequisites.

Available tools

Example prompts

• "Which Groq models can my key use right now, and what is the context window of each?"
• "Summarize this support email in three bullet points with llama-3.1-8b-instant and keep the reply under 200 tokens."
• "Transcribe the German podcast episode at https://example.com/audio/episode-12.mp3 with timestamps per segment."
• "Translate the Italian voice message at https://example.com/voice/note.m4a into English text."
• "Show me the status of my Groq batch jobs and how many requests failed in the last one."

Next steps

• Connect to Claude
• Connect to ChatGPT