An AI assistant connected to Speak-Y over MCP reads two things: your own recordings, and the recordings in the team channels you are a member of. It reads them from the library on your Mac. Nothing it reads becomes visible to your teammates, and nothing lands in a channel because you asked a question.
Publishing is a different operation with its own tool, share_to_channel. It
runs inside the Speak-Y app, the assistant is instructed to confirm the
recording and the channel with you first, and your MCP client asks for approval
before the call. This guide shows where the line between reading and publishing
runs, walks through one real request — "sort yesterday's meetings into
channels" — and lists what an assistant cannot do in a team workspace at all.
Everything here was checked against the macOS app on October 10, 2026.
The same recordings you see in the app, minus the audio. The Speak-Y MCP server joins two sources:
For each recording the assistant gets the transcript with speaker names, the summary, the action items, the tags and the channel it lives in. That makes a channel work as team memory: "what did we agree with the client about the deadline" is answered from the meeting a colleague recorded, not only from yours.
Two limits are built in rather than configured:
The assistant reads the copy of the channel that is on your Mac. If the answer
looks stale, ask it to sync: the sync tool pulls new recordings from your
other devices and from channel feeds, and it needs the app running.

Everything you have not shared. A team workspace does not turn your library into a shared drive, and the assistant does not change that:
One caveat applies to every MCP server, local or cloud: what the assistant reads is sent to your model provider as part of the conversation. Asking about a channel means text from that channel — including what teammates said — leaves the end-to-end encrypted space for that one conversation; see local vs cloud MCP. For a channel with hiring or board material, decide as a team whether assistants should be pointed at it.
Because it is the one assistant action that other people see. Reading stays on your machine; a recording shared to a channel is delivered to every member of that channel. Speak-Y keeps the two apart in four ways:
share_to_channel puts a recording into a channel, and it takes one
recording and one named channel per call.The work itself is done by the running app, signed in to your account: it holds the encryption keys, so the assistant never handles them. If Speak-Y is closed, the tool says so, and reading continues to work.

Take a workspace with a public channel Meetings and a private channel Product. Yesterday produced three meetings; one is already in a channel, two are not where they should be. The request to the assistant is one sentence.
list_recordings), opens the two that need a decision (get_transcript)
and asks the app for your channels (list_channels).launch, Speaker 3 is Maya Chen. Pricing call → move from Meetings to
private Product, tag pricing. Channel members will see them. Go ahead?"rename_speaker,
set_tags twice, then share_to_channel for the launch plan and
unshare_from_channel followed by share_to_channel for the pricing call.Two details of the order matter. Names and tags go first, publishing second: the channel receives the recording as it is at the moment of sharing, and a speaker renamed afterwards stays "Speaker 3" for the team. And a move is two calls, because a recording lives in exactly one channel: it is taken out of the old one, then shared to the new one. Between the two calls it is in no channel, so if the second one fails, share it again by hand with To channel — the route described in how to share a meeting to a team channel.

The list is short and deliberate:
create_channel tool exists for
the case "these interviews need a place that does not exist yet", and its
description requires asking you every time. A public channel can only be
created by the owner or an admin; any member may create a private one.Three settings, from narrow to broad:
share_to_channel,
unshare_from_channel and create_channel on "ask". An "always allow" on a
tool that publishes to colleagues removes the one human check in the chain.--read-only in that
client's config and the changing tools are not offered to it at all — a
reasonable default for an editor plugin you only use for questions.And read the plan before you say yes. The assistant picks a channel from a transcript and a channel name, so it can file a pricing discussion into a public channel as readily as a person in a hurry. Which meetings belong where is a decision the team makes once — see who sees what in meeting channels.

Most keep the assistant on the reading side. Checked on October 10, 2026:
All three run in the vendor's cloud, which has a practical advantage: they answer while your laptop is closed. Speak-Y's reading is local and its changes need the app, because only the app holds the keys of an end-to-end encrypted channel. If a server-side agent must reach team notes with no Mac online, a cloud note-taker fits better.
The Speak-Y MCP server is included in every plan, Free too. Team workspaces live in the macOS app: creating one starts on the Pro plan, and teammates join free on any plan. The Speak-Y Teams page shows how shared channels work, and what makes MCP write access safe goes deeper into approvals, logs and the two off switches.
Yes, if you are a member of that channel. The Speak-Y MCP server reads your own recordings and the recordings in the team channels you can open, from the copy on your Mac. Teammates' recordings are read-only for the assistant: it cannot tag, rename or move them.
It can call the share_to_channel tool, but not silently. The tool tells the assistant to confirm the recording and the channel with you, it is marked as an action that reaches other people so the client asks for approval, and every call is listed under Recent assistant actions in Settings → Integrations.
No. Reading and publishing are separate tools. Searching, listing and opening transcripts change nothing and send nothing to a channel; a recording reaches a channel only through share_to_channel, which runs in the Speak-Y app after you agree.
No. A channel's recordings reach your Mac only after they are decrypted with the channel key, and that key is issued to members. A private channel you were not added to is not in the data the MCP server reads, and it does not appear in list_channels.
Turn off Assistant actions in Settings → Integrations: reading keeps working, changes are refused for every client. To restrict one client only, start the server with --read-only in that client's config; the changing tools then disappear from its tool list.