Most platforms leave an old recording on whichever model processed it the day it landed. Here the analysis is re-runnable on current models, the transcript is editable in place, and search covers transcripts, insights and metadata at once — so the library answers better next year than it does today.
Speak AI's own server over its transcription and analysis workspace: interviews, sales calls, research sessions, webinars, podcasts and meetings, with 113 tools covering upload, transcripts, insights, search, clips, dashboards, automations and a meeting assistant. The recordings stay in the workspace; the client queries them.
- Media in by three routes — a pre-signed S3 URL for a local file, a public file URL, or a social link from YouTube, Instagram, TikTok, X, Facebook, Reddit or SoundCloud resolved for you — then polled from pending to processed
- Transcripts with speaker labels and timestamps, editable by find-and-replace, with speaker names renamed per file or across as many as 500 files in one call
- AI insights per recording — topics, sentiment, summaries, action items — re-runnable on newer models without re-uploading anything
- Deep search across transcripts, insights and metadata, and a chat that answers over one file, one folder or the whole workspace
- Captions as their own artefact, and exports as PDF, DOCX, SRT, VTT, TXT or CSV, singly or batched into one merged file
- Highlight clips cut from time ranges across files, and an embeddable player with an iframe URL for your own site
- A meeting assistant that joins a Zoom, Google Meet or MS Teams call, plus a live transcript call that returns only the sentences added since your last one
- Recorders and surveys with shareable URLs, folders and saved views, custom fields, user groups, analytics dashboards, and automations built from a trigger and a step graph
- Five resources readable without a tool call — the media library, folders, languages, and a given file's transcript or insights — and three prompts that run the multi-step jobs end to end
A Speak AI account. The hosted server is at https://api.speakai.co/v1/mcp over streamable HTTP: paste that into a client's connector settings and approve the OAuth consent screen, or send an API key as an Authorization: Bearer header for clients that do not speak OAuth. For stdio clients, npm ships @speakai/mcp-server, which takes SPEAK_API_KEY and an optional SPEAK_BASE_URL, and carries a 30-command CLI alongside it.
One command plus a key — npm install -g @speakai/mcp-server, then supply credentials