Edit Take skill

Your Task

Input: $ARGUMENTS

Change part of one take through Suno's MCP server and keep the rest of it: build the edit, show what it costs, get a go-ahead, send it, log the two edited takes, and hand off to listening.

You never choose a take, mark a keeper or set a track Final. The user listens and decides.

---

Edit Take Skill

An edit is the cheap fix for a take that is nearly right. Suno keeps the take's audio and changes only what the instruction names. Measured on a one-minute song: the edited takes had the source's exact length, sang the new word, and left the rest as it was. It is still a generation: one call makes two edited takes and costs 10 song credits.

Two ways to say what changes:

• In words — "drop the synth", "redo 0:45 to 1:00 with more energy", "change 'the light is gone' to 'the light came home'".
• From the track file — no words given: the edit brings the take in line with the Lyrics Box as it stands now. This is how a pronunciation or wording fix made in the track file reaches a take.

Past five changed places Suno re-sings the whole song as a cover; from there /bitwize-music:generate is the better tool. Background: ${CLAUDE_PLUGIN_ROOT}/reference/suno/mcp-api-mapping.md § Editing a Take.

Step 1: Resolve

Expected format: <album-name> <track> [take N] ["what to change" | instrumental]

Find the album and track as /bitwize-music:generate does. The word instrumental in place of an instruction asks for a version of the take, not a change to it: go to An Instrumental Version. If the words to change include new sung words, treat them like any new lyric: run check_homographs on them and ask about each homograph before going on.

Step 2: Check the Suno Server

Follow § Preflight in the reference doc: find audio-create, audio-wait, audio-retrieve and account-usage, load their current definitions, and read account-usage. Not connected → say an edit needs Suno's MCP server and stop; on suno.com the same edit is made in the song's editor.

Step 3: Build the Edit

• In words: get_edit_request(album_slug, track_slug, take: N, instruction: "<the user's words>").
• From the track file: call get_edit_request(album_slug, track_slug, take: N) once to learn the take's source.clip_id, then audio-retrieve(id: <that clip id>), then get_edit_request(album_slug, track_slug, take: N, take_lyrics: <metadata.lyrics>).

Leave take at 0 when the user named none: the tool picks the keeper, or the only take there is.

Result	Do
found: false	Report the error and stop
take_ambiguous	Ask which take, then call again with take
needs_instruction on the first call of the track-file path	Expected; carry on with audio-retrieve
nothing_to_change	The Lyrics Box already matches the take; ask what to change
Any other blocking problem	Show it and stop
edit_rewrites_song (a warning)	Say that Suno will re-sing the whole song, and ask whether to edit anyway or regenerate

Step 4: Show the Summary and Get a Go-Ahead

READY TO EDIT — [NN] [Title], take 2

  Change     "Waiting for the day to hold still"
        to   "Waiting for the tide to hold still"
  Sent as    In [Title], change "…" to "…".
  Keeps      the rest of the take's audio
  Makes      2 edited takes
  Cost       about 10 song credits. You have 9,915.
  Then       open both edited takes (free) and check them.

Go ahead?   yes · cancel

• In words mode, show the instruction in place of the Change lines.
• Wait for an explicit yes. Silence is not a yes.

Step 5: Send It

1. Fix an idempotency key before the first attempt: {album_slug}/{track_slug}/edit/{UTC timestamp}.
2. Call audio-create with the keys of request and the idempotency_key. Add nothing: duration, instrumental and vocal_gender make Suno regenerate the whole song, and lyrics and style are refused next to a description.
3. The answer carries two ids in clip_ids. Call audio-wait with both and until: "complete"; call it again while either is pending. An edit usually finishes within a minute.
4. Call account-usage again. Credits spent = balance before − balance after.

Failures are handled as in /bitwize-music:generate Step 6: one retry with the same key after a timeout; a new key if an argument had to be dropped; 402 is out of credits; 403 means the song may not be remixed or the plan lacks a feature; 400 is Suno's content check or a mixed-mode request. Never send a second call for one go-ahead except as that table says.

Step 6: Log and Hand Off

1. log_generation(album_slug, track_slug, <the track's model>, clips, notes: "edit of take N: <a few words>, via MCP, 10 credits"), with each clip's id, status and length.

2. Report:

   EDITED — [NN] [Title], take 2
   
     Take 5 (row 5)   0:55   https://suno.com/song/...
     Take 6 (row 6)   0:55   https://suno.com/song/...
   
     Spent 10 song credits. 9,905 left.
   

3. Continue into /bitwize-music:listen for the two new rows, then /bitwize-music:check-take for them, as /bitwize-music:generate Step 8 does.

The track's status does not change, and the original take stays in the log. If the user keeps an edited take, approval works as always (/bitwize-music:listen Step 7). When the edit was given in words and changed sung words, update the Lyrics Box and Streaming Lyrics to what the keeper sings.

An Instrumental Version

/bitwize-music:edit-take <album> <track> [take N] instrumental takes the vocals off a take and keeps the rest — a version to release beside the song, or to put under a video.

1. Step 2 as above, then get_edit_request(album_slug, track_slug, take: N, version: "instrumental"). already_instrumental, source_is_a_version and version_unsupported are blocking: show the problem and stop.
2. Summary, go-ahead and sending are Steps 4 and 5: 10 song credits for two takes. In the summary write Makes 2 instrumental versions of take N and Keeps everything but the vocals.
3. Log with notes: "<log_notes from the tool>, via MCP, 10 credits". Those words mark the rows as versions: they are never taken for the track's keeper, the take check leaves them out, and the track's status does not change.
4. Open both (/bitwize-music:listen <album> <track> take <rows>). Measured: both came back at the source's exact length with no voice on them. A version whose length differs from the source's was made again, not lifted out of the take — say so when it happens.
5. When the user picks one, mark that row ✓ and do nothing else: no Suno Link, no status change, no like. Then /bitwize-music:import-audio --from-suno <album> <track> --version instrumental saves it as versions/<track>-instrumental.wav.

An a cappella is not offered this way. Measured: the same request for an a cappella returned the instrumental once more and a shorter, re-sung song with its guitar. The vocals stem is the a cappella: /bitwize-music:import-audio --from-suno <album> <track> --stems saves it as stems/<track>/0 Lead Vocals.wav.

Remember

1. An edit is a generation: cost first, then a go-ahead, then one call.
2. Send only what get_edit_request returned, plus the idempotency key.
3. Two edited takes come back; the original is untouched.
4. More than five changed places is a re-sing, not an edit — say so before spending.
5. You never pick a take, mark a keeper or set Final.
6. A version is kept beside the track, never as it: its ✓ marks the one to import and changes nothing else.