Muse Code CLI Headless Exec with PilotCut
Run muse exec with META_API_KEY against PilotCut's local MCP for batch cleanup on a .pcbundle. Approvals, sandbox flags, and MCP not-sandboxed warnings from Meta docs. BYO only.
Muse Code CLI Headless Exec with PilotCut
The short version
Interactive Muse Code is muse in a project. For scripts and batch cleanup, Meta documents headless runs: muse exec "<prompt>" with META_API_KEY in the environment (auth, headless).
Point that agent at PilotCut's local MCP so batch work mutates a real .pcbundle: 83 purpose-built editing tools, shaped by tens of thousands of hours of video-editing practice, not Muse AI consumer tasks.
Muse Code remains BYO via MCP/CLI. It is not a PilotCut in-app catalog agent. PilotCut's featured agents are Claude Code, ChatGPT Codex, Grok Build, GitHub Copilot, OpenCode, Antigravity.
Free during beta. Bring the agent you already pay for. 83 purpose-built editing tools, shaped by tens of thousands of hours of video-editing practice. The fastest editor for AI: 83 purpose-built tools, shaped by tens of thousands of hours of video-editing practice, so your agent and you waste less time and money. Try PilotCut.
Install once, then exec
curl -fsSL https://dev.meta.ai/install.sh | sh muse --version
Interactive auth once (browser or API key), or always supply the key for automation:
export META_API_KEY="<your-key>"
Per Meta, an environment API key takes priority over a stored browser session (auth).
Wire PilotCut MCP before unattended runs
Declare PilotCut in Muse mcp_servers (stdio or streamable_http) while the PilotCut editor window that owns the .pcbundle is open (MCP).
MCP tools are not sandboxed (Meta). For headless runs, also read Meta's approval flags: --disable-approval keeps the OS sandbox but skips approval prompts; --yolo disables approval and the sandbox and trusts the workspace. Meta says use --yolo only on trusted code in a disposable, isolated container (headless).
PilotCut's Core Agent chat panel gates on riskTier. Muse Code as a third-party MCP client does not get that gate. Prefer prompts that demand confirm:false plans first on silence/filler tools, then an explicit second step for confirm:true.
Example headless shapes
META_API_KEY="$MUSE_KEY" muse exec \ "On the open PilotCut .pcbundle, run edit-words cutSilences and removeDeadAir with confirm:false only. Print the plan. Do not apply."
META_API_KEY="$MUSE_KEY" muse exec --json \ "Apply the reviewed silence plan with confirm:true, then export type=fcpxml to the path I specify."
muse exec --prompt-file ./pilotcut-batch-task.txt
Meta's exit codes reflect how the run ended, not whether the edit is correct (0 turn complete, 1 fail/cancel, 2 usage error). Gate on your own checks after the run (headless).
Optional: --max-model-steps caps a stuck loop. Resume with muse exec --session-id <uuid> "Continue." when you need a follow-up turn.
What batch cleanup is good for
- Propose breath-safe silence / filler plans across talking-head takes
- Re-run the same prompt file after human review of the plan
- Write
export type=fcpxmlwhen the next stop is Final Cut
It is not text-to-video. It is not Muse AI. It is a coding agent calling PilotCut tools on local media.
Eligibility and pricing
Meta positions Muse Spark 1.2 with expanded global access in Muse Code / Model API. Verify account eligibility at auth. Check Meta's live pricing pages for dollars. PilotCut stays free during beta with no editor-side credit meter for BYO agent edits.
Facts (checked 2026-09-15)
| Item | Value |
|---|---|
| Headless | muse exec "..."; CI via META_API_KEY |
| Approvals | --disable-approval vs --yolo (see Meta docs) |
| MCP | Not sandboxed |
| PilotCut | Free during beta; BYO agent; 83 tools; export type=fcpxml |
| Featured agents | Claude Code, ChatGPT Codex, Grok Build, GitHub Copilot, OpenCode, Antigravity |
| Muse Code | BYO only, not catalog |
Also read
- How to use Muse Code with PilotCut MCP
- How to use Claude Code / ChatGPT Codex MCP/CLI without Resolve Studio
- Edit videos with AI agents
Bottom line
Use muse exec + META_API_KEY for batch propose/apply loops on a .pcbundle, with Meta's sandbox and MCP warnings respected. Muse Code stays BYO outside the featured catalog list. Keep the cut on PilotCut's local timeline.
Try PilotCut. Free during beta. Bring the agent you already pay for.
Related articles
September 15, 2026
How to Use Muse Code with PilotCut MCP
BYO Meta Muse Code into PilotCut's local MCP: free during beta, confirm protocol on silence cuts, breath-safe snap, export type=fcpxml. Muse Code is not an in-app catalog agent.
4 min read
September 8, 2026
How to Use Claude / Codex with MCP/CLI for Video Editing (Without Resolve Studio)
Searching for Resolve 21.1 MCP setup with Claude or ChatGPT Codex? Here is the Studio path in brief — then the PilotCut path: BYO agent via terminal MCP/CLI on a local Mac project, free during beta.
5 min read
April 29, 2026
How to Edit Videos with AI Agents
A practical way to think about agent-assisted video editing: use AI for structure and repetition, then keep human taste in the final pass.
1 min read