Streamloop MCP

Find streams and their ids — the starting point for every other stream tool. Each row carries the name, current state, scheduled start and whether that schedule repeats, which is enough to tell a one-off from a daily loop without a second call. Filter by name substring, status (active = live or starting, scheduled = waiting for its window, inactive = neither) and creation window. Use get_stream fo

One stream's full configuration: current state, output quality and framerate, the destinations it publishes to (each enabled or not, with its delivery state), and its scheduled window including recurrence and repeat weekdays. Use it before changing a stream. For whether it could start right now, call check_stream_readiness instead (this tool reports what is configured, not what is missing); for li

Start streaming now, looping the stream's playlist (or playing its scene) until it is stopped. THIS PUTS VIDEO IN FRONT OF THE PUBLIC, so confirm with the user first unless they just asked for it. Requires a destination plus either a playable video playlist (at least one item) or an attached published scene (set_stream_scene) — check_stream_readiness reports what is missing without changing anythi

Stop a stream that is live or starting. It stops publishing and becomes inactive; nothing is deleted and it can be started again with the same configuration. Does NOT cancel a recurring schedule — a daily or weekly stream stopped here starts again at its next window, so call cancel_stream_schedule as well to keep it off.

Permanently delete a stream, its playlists and its schedule. Irreversible, and it cannot be undone by re-creating the stream: uploads survive but the arrangement does not. Stop the stream first if it is live. To keep a stream but take it off the air, use stop_stream (plus cancel_stream_schedule). Requires confirm=true and the streamloop:destructive scope.

Use this to set up a new loop or a stream for a scene. The stream is created empty and CANNOT start until it has (1) a destination — add_stream_destination with one from list_destinations, or get_connection_hints for how to make one — plus (2) either a playable video playlist (at least one item) or an attached published scene (set_stream_scene). Creating a stream also creates its two empty playlis

Change a stream's settings (name, quality, framerate, duration, storage, backup ingest) and/or its YouTube broadcast settings via the youtube object. Where it publishes is add_stream_destination / set_stream_destination_enabled / remove_stream_destination; when it runs is schedule_stream. The youtube settings only apply when one of the stream's destinations is a YouTube channel connected throu

Set the window a stream runs in, once or on a repeat, and arm it. The stream must already be startable — a destination plus either a playable video playlist or an attached published scene (check_stream_readiness) — otherwise the window arrives and the start fails. repeat=daily re-arms the window exactly 24h later; repeat=weekly re-arms +7 days, or on the weekdays given in days. Both repeats REQU

Disarm a scheduled stream and return it to an unscheduled, inactive state, clearing any daily or weekly repeat. Use this to stop a recurring stream coming back; it has no effect on a stream that is currently live (stop_stream does that).

Answer 'would start_stream work right now?' without changing anything. Runs the same pre-flight configuration checks a start does and returns ready plus every blocking issue — typically a missing or archived destination, an empty video playlist, media too big for the stream's storage, or, for a stream playing a scene, a scene that is missing, not published or invalid. It does NOT check whether upl

How a live stream is doing ON YOUTUBE: watch URL, broadcast status, concurrent viewers, views and likes. Only meaningful while the stream is live to a YouTube channel connected through Streamloop — a custom RTMP destination reports nothing here, since those numbers live with that platform. For whether Streamloop itself considers the stream healthy, read state from get_stream.

Where a stream publishes, in the order they were added: each destination, whether it is enabled, and while the stream runs its delivery state (connecting, live, reconnecting, failed with a reason).

Make a stream also publish to one of the workspace's destinations (list_destinations), enabled. Up to 5 per stream and one YouTube channel at most (DESTINATION_LIMIT, YOUTUBE_DESTINATION_LIMIT, DESTINATION_DUPLICATE). A second destination needs multistream on the workspace (MULTISTREAM_NOT_ENABLED). Refused while the stream is live (STREAM_LIVE), and for a destination the stream's quality or codec

Stop or resume sending to one of a stream's destinations without removing it. Disabling works while the stream is live — that output goes off the air at once — except for its last enabled destination (LAST_DESTINATION: stop_stream instead). Enabling is refused while live (STREAM_LIVE).

Take a destination off a stream (the destination itself stays in the workspace). Refused while the stream is live (STREAM_LIVE): disable it with set_stream_destination_enabled instead.

Use this FIRST when the user names a platform they want to stream to ("put this on YouTube", "also send it to Twitch") and there is no destination for it yet. Returns, for that platform: the connection methods, the exact ingestUrl to pass to create_rtmp_destination (or null when the platform issues one per session), what each method can and cannot do afterwards, the Streamloop tools to call in o

Find the places this workspace can publish to — connected YouTube channels and custom RTMP endpoints — and their ids, which is what add_stream_destination needs. Check status: only an active destination can carry a stream; needs_reauth and revoked mean the YouTube connection must be re-consented (start_youtube_connection) before a stream using it will start. Filter by name substring, a sin

Create a destination for any platform that gives out an RTMP ingest URL and stream key (Twitch, Facebook, Kick, X, TikTok, Telegram, a self-hosted server, or YouTube when its channel is not connected here). Take rtmpUrl from get_connection_hints(platform).ingestUrl rather than guessing it — a wrong host is a stream that silently never connects. The key is stored encrypted and never read back. Use

Maintain an existing destination: rename it, replace the stream key after the platform rotated it (custom RTMP only — a connected YouTube channel has no key to set), and archive or restore it. Archiving hides a destination from the picker without deleting it; it is reversible with archived=false. A live stream keeps publishing with the key it started with, so restart the stream after rotating a ke

Begin connecting a Google account so its YouTube channels can be streamed to with managed broadcasts (titles, privacy, thumbnails, live stats). Returns a Google consent URL — give it to the user to open in a browser; consent cannot be completed by an agent. Afterwards, list_youtube_channels shows the authorized channels and create_youtube_destination turns one into a destination. Also the fix for

The YouTube channels this workspace has Google consent for, with live channel data (title, handle, country, subscriber count, connection status). These are NOT destinations yet: pass a channel's id to create_youtube_destination to get something a stream can publish to. Use list_destinations to see which channels already have one. An empty result means no Google account is connected — start_youtu

No parameter schema on file.

Turn an already-consented YouTube channel into a destination a stream can publish to, with managed broadcasts (update_stream's youtube settings, get_stream_live_stats). Takes the channel id from list_youtube_channels — NOT the YouTube channel URL or handle. If that list is empty, the account has not consented yet: start_youtube_connection first.

Use this to get from a stream to something you can edit: pass streamId and you get the two playlists that stream owns — one video, one audio, created empty with it. Pass sceneId for the playlists a scene's sources play. Also the way to find a playlist id before any other playlist tool. Rows are a summary; get_playlist has the items.

An empty playlist owned by a scene, for one of its playlist sources to play (a stream's own two playlists come with the stream). Fill it with apply_playlist_operations. Each call creates another.

Use this before editing a playlist (item ids for remove and reorder come from here, and so does the version the edit tools take) and after swapping one onto a live stream (to see the swap finished). Returns the draft items in play order, what a running stream is playing right now, whether the draft differs from that, how it differs, and any custom playback order attached.

Use this for every playlist change: add media, remove items, reorder, or switch between sequential and shuffle, as one ordered batch. All-or-nothing — if any operation is invalid, nothing is written. A stream that is NOT currently live picks the change up on its next start, with nothing else to call. A stream that IS live keeps playing the arrangement it started with until swap_live_playlist repla

Use this to reuse an arrangement on a second stream instead of rebuilding it: copies one playlist's contents into another playlist of the same kind. The target must be EMPTY — this never merges or overwrites. The target stream picks the contents up on its next start; if it is already live, swap_live_playlist puts them on the air.

Use this ONLY to put an edited playlist on the air while its stream is LIVE — the hot swap the product calls publishing. The stream changes over to the new arrangement without going off the air. NOT needed to set a stream up and NOT needed before a first start: starting a stream takes the playlist as it is then, so on a stream that is not live this tool just fails with NOT_LIVE. Asynchronous — new

Find saved custom playback orders (order sequencers) and their ids. Use it before creating one — an order the user already has can be assigned to another playlist instead of rewritten. Pass streamId to see only orders that can run over that stream's playlist. latestReadyVersionID null means nothing usable has been produced yet. Which order a given playlist is using is on get_playlist, not here.

One saved order: its description, how many playlists use it, and its newest usable version with any validation problems. Pass playlistId as well to get the answer that actually matters before assigning — whether this order can run over THAT playlist (assignable), which items it expects but cannot find, and a dry-run of the first items it would pick, in order. Do that instead of assigning to find

Describe a playback order in plain language and have one written and validated. Pass the playlist it is for as playlistId: the order is written against that playlist's REAL items, so leaving it out produces a rule aimed at nothing. Creating does NOT apply it — assign_playlist_order does. Read status on the result: READY = usable; INVALID = written but it failed validation, and diagnostics say

Continue the conversation about an existing order — 'never end on an instrumental', 'put the announcement second instead' — producing a new version. Use this rather than creating a second order for the same purpose. Pass the playlist currently being composed against so the rewrite sees its real items. Interpret status exactly as for create_playlist_order. A new version does NOT reach playlists a

Make a playlist play in a saved order. This is the step that changes playback, and unlike a playlist edit it needs no swap_live_playlist — a running stream picks the new order up for its next items. The version in use is pinned at assign time and never advances on its own, so call this again after refine_playlist_order to adopt an improvement. Check get_playlist_order(id, playlistId) first: assign

Stop a playlist using a custom order; it returns to playing in its own item order (or shuffled, per the playlist's mode). The order itself is kept and can be assigned again. Like assigning, it takes effect on a running stream immediately, with no swap_live_playlist needed.

Find media and their ids (obj_...) — what apply_playlist_operations adds to a playlist. Removed media is not listed. uploaded false means the media is still arriving or processing and cannot be streamed yet. Filter by name substring, kind and creation window; use get_upload for one item's duration, resolution and import progress.

One media file in detail: duration, resolution, framerate, whether it has video and audio (probed from the file itself, not from its name), a temporary download URL, a thumbnail, and — for a URL import — how far the download has got and why it failed. Use it to confirm media is ready to stream, and to read the duration a schedule window has to fit.

Step 1 of uploading a LOCAL file: reserves the media record and returns uploadUrl, a presigned PUT you can use immediately. This server does not relay bytes, but YOU probably can — if the client can run a command or make an HTTP request, do the PUT itself: curl -X PUT -H 'Content-Type: <the mimeType you passed>' --data-binary @<file> '<uploadUrl>' The Content-Type header MUST equal the mimeTyp

Step 2 of uploading a local file: call it once the bytes have been PUT to the url create_upload returned — by you, or by the user — which starts processing. Until this is called the media is unusable, and calling it before the PUT finishes leaves a broken record. Nothing to call for a url import: import_upload_from_url completes on its own.

Look at any media url and report what could be imported from it: each asset with its quality options, sizes and duration. REQUIRED BEFORE import_upload_from_url, which takes the opaque asset and quality ids only this tool hands out — they cannot be constructed. Works for a media page AND for a direct file link, which is what makes it the right tool for a FILE THE USER ATTACHED in this conversation

Import chosen assets from a media url — a page, a direct file link, or a file the user attached in this conversation — straight into the workspace, with no local file and no PUT. Each selection must echo an assetId and qualityId from discover_external_media for the SAME url; invented ids are rejected. One selection produces one media item, so the same asset can be imported twice with different ran

Permanently remove a media file from the library: its stored bytes and every file derived from it are deleted. Irreversible. An upload that never finished (no complete_upload after the PUT, an import that failed) is dropped too: nothing was stored. Refused, with nothing changed, while a playlist or a recent stream still uses it (IN_USE) — remove it from those playlists first with apply_playlist_op

Use this to answer 'can this keep streaming, and for how long' — and whenever a stream fails to start or stops with no configuration problem, since check_stream_readiness deliberately does not look at credits. One call returns the credit balance, credits left, whether the account is in good standing, how fast credits are being spent right now, which streams are live and what they are costing, and

No parameter schema on file.

Use this for questions about the past — 'how many hours did I stream last month', 'is my spend going up' — not about the present. Returns minutes streamed, credits spent and session counts bucketed over a period and split by output quality. get_billing_overview has the current balance and burn rate instead.

Who this token belongs to: the signed-in account's id, email, name, language and the default run-length new streams inherit. Use it to confirm which account an agent is acting for. It says nothing about which workspace is in use (get_selected_workspace) or about credits (get_billing_overview).

No parameter schema on file.

Every workspace this account belongs to, with the account's role in each and which one is currently in use. Streams, media, playlists, destinations and credits all belong to ONE workspace and are never shared between them, so when a user says something exists and it is not in a list, check here before concluding it is missing — the answer is usually that another workspace is selected. Pass an id

No parameter schema on file.

Switch which workspace this connection works in, for every later tool call on this connection (other connections, even ones sharing the same login, are unaffected). Takes an id from list_workspaces; membership is checked here and again by the API on every request, so this grants no access by itself. Use it when a stream, upload or destination the user describes is not in this workspace. Override

Which workspace this connection is working in right now, and where that came from: explicit (a select_workspace call), header or env (the client pinned it in its own configuration), or token-hint (the default the login carries). Call it when a result is unexpectedly empty, to confirm you are looking in the right workspace before telling the user something does not exist.

No parameter schema on file.

Read the scene's draft (or a published version, with version): a resource (with its revision and children, the sub-paths to read next), a type such as element/Text or source/data (its props), a mask ending in /* (a list), or "" (the types). as: "image" on a frame or layer → a picture of it. (In a scene's draft; scene: its id.)

Fetch a URL once, changing nothing in the show: HTTP status, content type, an outline of the JSON (keys, their values, array lengths) or the page's title and text, and the rows a table would get from it with connector, pick and headers. Use it before creating a URL table, and to find the pick path or selector. (In a scene's draft; scene: its id.)

Delete whole resources (frames, layers, sources, components). All or nothing: anything in use by something not removed in the same call is refused, with what uses it. (In a scene's draft; scene: its id.) Removing a resource needs its revision in revisions (path → the revision you read; "*": whatever is there, on purpose); without it the answer is REVISION_REQUIRED.

Create, replace or change one resource; get <type> shows what each takes, with an example. value is the whole resource, as get answers it: a frame { code: "<JSX>" } or, as JSON, { name, root }; a component { description, props, tsx }; a source or table its fields. edits change part of it: [{ old, new }] replaces text in its code (copy old from what get answers: code is stored formatted), or an obj

Lists the workspace's scenes, newest first: id, name, the version last published (0: never) and when. Read one's draft with scene_get (path "" lists what it holds).

Creates a scene with an empty draft, nothing published. Build it with scene_set (frame/<id> for each frame, source/… for what it reads), then publish_scene. Each call creates another scene.

Publishes the draft exactly as it was at draftRevision (required) as the scene's next version: what streams playing it play from then on. All of it is checked first; a refusal (SCENE_INVALID) lists each problem — fix them with scene_set and publish again. Edits made after draftRevision are not published (the answer says so, stale: true). If someone published after the version your draft is based o

Puts the draft back to the published version: every change made since it was published is lost (nothing on air changes). draftRevision (required) is the draft you reviewed: if it changed after that, nothing is discarded (CONFLICT). dryRun: true lists what would be lost and discards nothing — show the user that first. Refused (NOT_PUBLISHED) when nothing is published.

Lists a scene's published versions, newest first: number, when, by whom. Read one with scene_get and version.

Publishes an earlier version's document as the next version (the versions in between stay; the draft is untouched — publishing the draft later is refused with STALE_PUBLISH until you send overwritePublished, and discard_scene_draft makes the draft this version too). expectedVersion (required) is the latest published version you saw (list_scene_versions, list_scenes): if someone published since, no

Deletes the scene, its draft and every published version, for good. Refused (SCENE_IN_USE) while a stream plays it. Its ingest keys are revoked and its cameras removed.

Makes a stream play a scene (its published version) instead of a video playlist, or with scene null stops doing so. Refused (STREAM_LIVE) while the stream runs: stop it, set the scene, start it again. The scene must be published first (publish_scene).

Answers each stream that plays it: the stream's state and, under onAir while it runs, the frame/<id> on air (frame) and the one cued next, the controls' values, each source's status and what its playlists play.

Acts on the scene on air, as its operator would: goFrame (take frame/<target> to air, with a transition), next (cue frame/<target> so the take is instant; no target clears the cue), setControl (set one of its controls; scene_get control/* lists them), setData (replace a table's rows), skip (a playlist source to its next item). Viewers see it at once. The arguments are checked first, then against w

Makes the key an encoder (OBS, vMix, a phone app) sends a live feed with, for one of the scene's rtmp or srt sources: RTMP, RTMPS and SRT URLs with the key, shown in this answer only — give them to the user. Make the source first (scene_set source/rtmp/<id>). The source's status is waiting until an encoder connects with this key, and live while it publishes. The key's own id (ingkey_…) is what rev

Renames the scene (what the dashboard and list_scenes call it).

Lists a scene's ingest keys (id ingkey_…, which source, live or not, never the key itself), its secrets (names only) and its IP cameras: what create_scene_ingest_key and set_scene_secret made.

Revokes an ingest key: the encoder using it is cut off at once. A new key for the same source: create_scene_ingest_key.

Deletes a saved secret: a table that names it ("$secret:<name>") stops getting it.

Saves an API key or token a table sends (sealed; never shown again). A table names it as "$secret:<name>" in a header value or a URL's query. Only a key the user gave you.