bitrix24_call_transcribe
Transcribe Bitrix24 CRM call recordings locally and offline using Whisper large-v3. Provide a call activity ID to receive the transcript text, segments, manager ID, and call direction without any data leaving your machine.
Instructions
Transcribe a CRM call recording LOCALLY and fully offline (audio never leaves the machine). Resolves the call's recording (Voximplant/Mango RECORD_FILE_ID), downloads it, and decodes it with local Whisper large-v3 (strong Russian). Works out of the box — the MCP auto-bootstraps its own Python venv + model on first use (first call is slow: it downloads the ~3 GB model). A bounded worker pool serves calls, so firing hundreds at once just queues and drains N-at-a-time. This does NOT write anything to Bitrix — it only returns the transcript. To store it, the caller saves it with bitrix24_crm_timeline_note_save. Returns {text, segments, responsibleId, direction}: raw, unlabelled segments (Whisper splits by pause, not by speaker). For «who-said-what» labels the calling model decides — it has responsibleId (the manager) here and the client's name on the lead. Substitutes for Bitrix's UI-only call transcription/BitrixGPT scoring, which is NOT in the REST API. NOTE: phone audio is 8 kHz mono (mixed-speaker), so expect substance-accurate but imperfect text.
Tool call template
Replace each placeholder with a value matching the input schema. Only required parameters are included; add optional parameters as needed.
{
"id": 1,
"jsonrpc": "2.0",
"method": "tools/call",
"params": {
"arguments": {
"activityId": "<activityId>"
},
"name": "bitrix24_call_transcribe"
}
}Input Schema
| Name | Required | Description | Default |
|---|---|---|---|
| activityId | Yes | The call activity ID (from bitrix24_crm_activity_list — a VOXIMPLANT_CALL activity) |