Claude Video Editing Workflow Guide

Claude's strength is not dragging clips on a timeline — it is holding a plan across many tool calls without losing thread. That makes Claude a natural operator for video pipelines when paired with MCP-native infrastructure like VisionDraft.

This workflow guide walks through Claude video editing from Desktop setup to production patterns: uploads, captions, renders, and handoff to distribution. VisionDraft is the execution backend; Claude is the conductor.

Start with Claude MCP explained if you have not configured MCP yet.

Stack Components

Layer	Role
Claude Desktop	MCP host + user interface
VisionDraft MCP	Video tools at /api/mcp
Supabase	Projects, assets, jobs metadata
Render worker	FFmpeg processing (off Vercel)

Signup: /signup. MCP config: /mcp. Reference: /docs.

Phase 1: Configure Claude + VisionDraft

1. Obtain Server URL and vd_... API key from dashboard
2. Add VisionDraft to Claude Desktop MCP config with Bearer header
3. Restart Claude Desktop
4. New chat: "List VisionDraft tools and call list_projects"

Confirm tools appear: create_project, upload_asset, create_upload_url, complete_upload, list_assets, generate_captions, render_project, get_render_status, download_export.

Phase 2: Intake — Project + Media

Create project

Create a VisionDraft project "Interview — Jane Doe" and save the project ID in this chat.

Claude calls create_project. Store project.id in conversation context.

Upload strategy

File size	Tool path
Under ~4MB	upload_asset (base64)
Larger	create_upload_url → HTTP PUT → complete_upload

I have a 350MB MOV. Create a signed upload URL for interview-jane.mov in project {id}.

Claude returns URL and asset_id. Upload externally, then:

Complete upload for asset {asset_id} and list_assets to confirm.

Phase 3: Transform — Captions

Generate English captions for asset {asset_id} in project {id}. Tell me segment count when done.

Claude invokes generate_captions:

• Downloads asset to worker temp dir
• Runs Faster-Whisper transcription
• Persists caption record + timeline segments

Deep dive: generate captions using AI.

Phase 4: Produce — Render + Delivery

Render project {id} as jane-doe-interview-captioned with burn_captions true. Poll get_render_status every 30 seconds until status is completed or failed. Then download_export.

Expected chain:

1. render_project — auto-adds first video clip to timeline if empty
2. get_render_status loop
3. download_export — signed URL (typically 1-hour expiry)

Save URL to your DAM or share with client.

Phase 5: Review Loop

Humans review export. If fixes needed:

• Re-run generate_captions with different language
• New render_project with distinct export_name
• Compare exports in VisionDraft dashboard

Production Workflow Templates

Weekly show episode

1. create_project(name: "Show — {date}")
2. create_upload_url for raw recording
3. [human upload]
4. complete_upload
5. generate_captions language en
6. render_project export_name show-{date}-master
7. poll + download_export

Paste into Claude as a standing instruction for your team.

Multi-deliverable from one master

1. Master render (16:9 captioned)
2. Separate projects for shorts with different export names
3. Claude tracks multiple job_id values in a table in-chat

Claude vs. ChatGPT for Video

Factor	Claude	ChatGPT
MCP config	Desktop file, multi-server	Connector UI
Long tool chains	Strong	Strong
Ecosystem	Anthropic-first teams	OpenAI-first teams

See ChatGPT video editing guide for parallel setup.

Combining Other MCP Servers

Example same-session flow:

1. Filesystem MCP — list ~/recordings/*.mp4
2. VisionDraft — ingest + render
3. Slack MCP — post download_export link

This is future AI agent workflows in practice.

Quotas and Errors

VisionDraft enforces plan limits on pricing. Claude surfaces server errors — interpret and retry:

• No video assets — upload incomplete
• Quota exceeded — upgrade or wait
• Export not completed — keep polling

Why Infrastructure Matters

Claude without execution backend only describes edits. VisionDraft provides:

• Non-destructive timeline JSON
• Real FFmpeg output
• MCP tool schemas Claude reads at runtime

Not another AI video editor — infrastructure for agents.

Related: traditional vs AI agent editing, complete guide to AI video automation.

Parallel Renders in One Claude Thread

Claude can track multiple job_id values in a markdown table within the chat. Example:

Deliverable	project_id	job_id	status
Master 16:9	proj_1	job_a	completed
Short hook A	proj_2	job_b	queued

Ask Claude to update the table each poll cycle — human-readable ops dashboard without leaving the conversation.

Handoff to Non-Technical Stakeholders

After download_export, Claude drafts an email with:

• Link (note expiry time)
• One-paragraph summary from caption text
• Review checklist (names, dates, compliance claims)

Reduces friction between production and approval chains.

macOS vs Windows Claude Desktop Paths

Configuration file paths differ by OS. Document your team's exact path in internal wiki after first successful setup. Wrong-file edits are the top support issue for Claude MCP beginners.

Offline Recording, Online Render

Field teams record offline. Back at desk:

"Create project, guide me through create_upload_url for 2GB file."

Claude walks through signed URL upload without requiring stable chat during field recording.

Escalation From Claude to Engineering

When Claude hits repeated FFmpeg failures, escalate with job_id, project_id, and timestamp to platform support. Structured IDs beat screen recordings of timeline UIs.

Session Handoff Between Operators

End Claude thread with summary message:

"Save state: project_id X, asset_id Y, job_id Z pending."

Next shift operator pastes summary into new thread — reduces rediscovery. Better: store IDs in team Postgres regardless of chat.

Voice Input to Claude

Dictating workflow instructions via OS voice typing speeds field producers. Verify project names spelled correctly before create_project.

Export Expiry Alarms

Calendar reminder 45 minutes after download_export if file not yet copied to DAM — signed URL may expire at 60 minutes.

Collaborative Review in Claude

Share Claude conversation link internally (if org policy allows) so producer and reviewer see same tool trace — transparency beats screenshot ping-pong.

Long-Running Claude Sessions

Multi-hour renders may exceed chat context. Start fresh thread with pasted state block (project_id, pending job_id) — Claude recovers via tools better than memory.

Claude for Agencies

Agency uses one Claude Project per client with client-specific VisionDraft API keys in separate MCP server config entries named visiondraft-client-a. Prevents cross-client project leakage.

Billing Pass-Through

Agencies reselling automated video pass VisionDraft pricing render costs to clients with margin — document per-render unit cost in statements.

Pairing With Frame Review Tools

Reviewers comment on Frame.io link while Claude session holds MCP state — parallel workflows until integrated review MCP exists.

Reference Appendix: Implementation Notes

Production teams should treat this guide as a living document tied to VisionDraft's MCP tool surface at /docs. Before any batch automation goes live, run a golden path test on a five-second sample clip: create_project, ingest, generate_captions, render_project, poll get_render_status, and download_export. Archive the resulting job_id and export_id as regression fixtures.

Credential hygiene remains the top security issue. API keys from /mcp belong in host connector settings or secrets managers — never in blog comments, ticket attachments, or Git repositories. Rotate keys when employees leave or when a connector was exposed in a screen share. For agencies, separate keys per client prevent accidental cross-posting of exports between brands.

Quota planning on pricing avoids mid-campaign surprises. Model monthly demand: number of episodes × (caption minutes + render minutes per episode) + Shorts derivative factor. Upgrade tier before Black Friday or conference season, not after queue saturation. VisionDraft enforces limits server-side; agents surface errors but cannot override billing.

Async discipline separates hobby workflows from production. Every operator must internalize: render_project returns immediately; completion requires get_render_status polling until completed or failed. Scripts should use exponential backoff (30s, 45s, 60s caps) and alert if p95 latency exceeds SLA. Do not chain duplicate render calls hoping to "speed up" a stuck job — diagnose the existing job_id first.

Human review gates protect brand and compliance. Automate mechanical captioning and encoding; keep humans on claims, regulated statements, music rights, and talent releases. Download URLs from download_export expire — copy files to your CDN or DAM within the signed URL window (typically one hour).

Cross-host portability is a core benefit of MCP-native infrastructure. The same VisionDraft project namespace works from Claude Desktop, ChatGPT connectors, or headless JSON-RPC clients. If one host has an outage, failover procedures should document alternate host configuration hitting identical Server URL and a backup API key.

Observability: log project_id, asset_id, job_id, and export_id for every production run. When stakeholders ask "which export went live Tuesday?", IDs answer definitively unlike chat transcripts. Pair logs with VisionDraft dashboard render history during postmortems.

Related reading: what is MCP, complete guide to AI video automation, VisionDraft MCP infrastructure. Next step: create your account and configure /mcp to run the golden path test today.

Frequently Asked Questions

Is Claude good for video workflows?

Yes — strong multi-step MCP orchestration.

Need Claude Desktop?

Recommended for stable MCP configuration.

Long renders?

Poll get_render_status until complete.

Multiple projects?

Use list_projects and explicit project_id every call.

VisionDraft's role?

MCP-native cloud execution for video tools.

---

Run Claude on real video infrastructure. Create an account and connect via /mcp.