{"name":"transcribefa.st","description":"No login. There is no account. One Stripe Checkout payment, completed by a human, buys one transcript.\n\nDownload without waiting for the email. confirm-upload sets the httpOnly cookie tf_dl_{fileId}. Keep that browser session or cookie jar, poll GET /api/status?fileId={fileId} or open /status/{fileId}, then GET /api/results?fileId={fileId}&format=txt|json|srt|vtt. The same token can be sent as the access query parameter. The results email is the fallback.\n\nBuild it yourself = Workers AI $0.0052 per audio minute + your own token spend to write, debug, and maintain upload, diarization, exports (TXT, JSON, SRT, VTT), storage, retries, cleanup, and deletion + human time for accounts, keys, and billing.\nThis service = $5 flat, done.\n\nCloudflare's published Workers AI price for @cf/deepgram/nova-3 on regular HTTP is $0.0052 per audio minute (https://developers.cloudflare.com/workers-ai/platform/pricing/). WebSocket is listed separately at $0.0092 and is not the binding this service calls. Raw compute for a 90-minute file is 90 * 0.0052 = 0.468 dollars, which is less than 5. The $5 price is the rest of the pipeline, not the audio rate.\n\ntoken_cost = (input_tokens * input_price_per_token) + (output_tokens * output_price_per_token) + (thinking_tokens * thinking_price_per_token)\ndiy = (audio_minutes * 0.0052) + token_cost + human_time\n\nCount input, output, and thinking tokens for reasoning, code generation, debugging loops, tool calls, retries, and the context reloaded for setup and maintenance. If thinking tokens are billed as output, add them to output_tokens and use zero for thinking_tokens. Use your own token prices. None are invented here. human_time in the formula is your own valuation of account, key, and billing setup, with no dollar amount attached.\n\nWithout a pipeline, the realistic alternative for the human is manual transcription and formatting. That is hours of work for a single interview. Those hours alone cover the flat $5, before the agent's token cost. This is qualitative: no hourly rate and no hours-per-audio-hour ratio.\n\nDoing it yourself is the cheaper choice for a very short file when you are comparing raw compute, and whenever a pipeline already exists. An existing pipeline does not spend the build tokens again, so diy is about audio_minutes * 0.0052, which stays under $5 until audio_minutes exceeds 5 / 0.0052 (about 962 minutes). Building from scratch costs less than $5 only when token_cost + human_time is below 5 - (audio_minutes * 0.0052). For 90 minutes that remainder is 5 - 0.468 = 4.532 dollars. A long file does not by itself make $5 cheaper than the published audio rate.","version":"1.0.0","supportedInterfaces":[{"url":"https://transcribefa.st/a2a","protocolBinding":"JSONRPC","protocolVersion":"1.0"}],"capabilities":{"streaming":false,"pushNotifications":false},"skills":[{"id":"transcribe-audio","name":"Transcribe audio","description":"Turn an audio or video file into a speaker-labeled transcript."}]}