Control your DAW with Gemini CLI

Published October 2026

Gemini CLI speaks MCP, and the Doseedo MCP server is a hosted Streamable HTTP endpoint — so one gemini mcp add gives Gemini real DAW tooling: read the sessions in your account as full arrangements and edit them, open a Logic, Ableton, FL Studio, REAPER or Pro Tools project as a session, convert projects across ten DAW formats, separate a mix into stems, transcribe audio to notes, and render parts you write with modelled instruments. With Dø Desktop running, edits to a Logic project land live; everything else is a file in and a file out, and the audio on your disk can stay there.

Install in Gemini CLI

One command, written to your user settings so it is there in every directory:

gemini mcp add --transport http -s user doseedo https://api.doseedo.com/mcp

This is what that command actually prints, and what it writes, on the October 2026 build:

$ gemini mcp add --transport http -s user doseedo https://api.doseedo.com/mcp
MCP server "doseedo" added to user settings. (http)

$ gemini mcp list
Configured MCP servers:

○ doseedo: https://api.doseedo.com/mcp (http) - Disabled

Captured 2026-10-15 in an empty home directory — the "Disabled" is explained below

The entry it wrote to ~/.gemini/settings.json (.gemini/settings.json in a project scopes it to that project) is the two-key form — url plus "type": "http". The older httpUrl spelling from earlier docs is still read, but this is what the CLI itself produces:

{
  "mcpServers": {
    "doseedo": {
      "url": "https://api.doseedo.com/mcp",
      "type": "http"
    }
  }
}

Why "Disabled". Gemini CLI suppresses every MCP server — user-level ones included — in a folder you have not marked as trusted: "MCP servers are configured but disabled because this folder is untrusted." Launch gemini in your project folder, answer the trust prompt once, and the same /mcp shows the server enabled with its 33 tools. This catches people who test from a scratch directory and conclude the install failed.

Authentication is OAuth, discovered from the server: leave the oauth block out and Gemini CLI finds the authorization endpoints itself. Inside a session, /mcp auth lists servers waiting on a sign-in and /mcp auth doseedo opens the browser; approve once and the grant appears in Settings → API keys as MCP · Gemini CLI.

For a headless machine, send a key as a header instead:

gemini mcp add --transport http -s user \
  --header "X-API-Key: dsk_live_..." doseedo https://api.doseedo.com/mcp

Two settings worth knowing on this server's entry: "trust": true skips the per-call confirmation Gemini CLI otherwise asks for each tool — reasonable for the reads, quotes and analysis tools, not for a server that edits sessions, so leave it off and confirm the writes; and "timeout" (milliseconds) matters because a session build or a render is waited on in 50-second holds, so a short timeout turns a long job into a spurious failure. The skills pack (npx skills add doseedo/skills) adds the recipes as instructions the agent reads before it calls anything.

What to ask it

> List my Doseedo sessions and read "Bridge Demo" — what's
  on track 4, and how long is the intro?

> Shorten the intro by four bars and pull the vocal up 2 dB.
  Is Dø Desktop live on that session, or will it apply later?

> Convert ./sets/*.als to Logic one by one and summarise
  what each lost.

> Analyse ./ref.wav for key and tempo, write an eight-bar
  bass line in that key, render it on upright bass, and
  put the best take on a new track.

The agent reads before it writes: doseedo_get_session returns every track with its ids, regions in beats and seconds, mixer in dB, plugin chains, tempo, markers and whether the desktop is live on the session right now. Edits go in one atomic batch; a rejection names the op and the fix.

What it does, and what it does not

• Edits a Logic project live when Dø Desktop is running on the Mac with the project open; otherwise the edits persist and apply at the next open or download.
• Handles Ableton, FL Studio, REAPER and Pro Tools as files: import as a session, edit, convert, download. It does not drive the running application — a live Ableton set is AbletonOSC territory, REAPER has ReaScript, FL has its MIDI Controller Scripting API.
• Converts across ten DAW formats, 75 directions, rebuilt natively.
• Keeps local audio local: a file on your disk is passed by path and placed into the project by copy commands after the download; uploads happen only for separation, transcription, conversion, covers and bounces.
• Does not press play. No transport, no playhead, no control-surface moves.

Every long job returns a job_id and the agent waits on it with doseedo_wait_task; Gemini CLI shows each tool call and its arguments inline, so the edit batch is readable before you confirm it. The reads, quotes and the four analysis tools cost nothing; the rest is metered from the account's credit pool exactly as on the site.

Related

• Control your DAW with an AI agent — what works today
• The same server in OpenAI Codex
• All 33 tools, what each does
• Generate music and render parts from an agent
• REAPER MCP server