If you need multiple-language versions of the same video without repeating manual edits, Claude connected to Kapwing via the Model Context Protocol (MCP) gives you a repeatable, scriptable workflow for captioning, subtitle translation and AI dubbing. Instead of uploading and editing each clip by hand, Claude issues the steps and Kapwing’s tools execute them—useful for marketing teams and creators who want consistent outputs across many videos and languages.

What this setup does and when to use each workflow

The integration supports three distinct tasks. Choose according to whether you want the original voice preserved, readable translations, or fully localized audio:

  • Video captioning — Transcribes the original language and produces timed captions. Use it for accessibility and social platforms where viewers rely on text.
  • Subtitle translation — Translates captions into another language while keeping the original audio. Use it when you want native-language text but keep the speaker’s voice.
  • AI video dubbing — Translates dialogue and generates synthesized speech in the target language, producing a new audio track plus matching captions. Use it when viewers should hear the content in their language.

The MCP lets Claude coordinate Kapwing tools for uploading footage, transcribing dialogue, translating text, generating dubbed audio and exporting completed files. The approach is fastest on short, clear clips; multi-speaker or noisy recordings require more manual review.

Required components and initial setup

To run the flow you need:

  • A Claude client (Claude desktop, Claude Code or Codex) to send commands.
  • A Kapwing account to host projects and exports.
  • The Kapwing MCP connector installed and authenticated in Claude.
  • The three Kapwing skills for this flow: /captions, /translate and /dubbing.
  • The original video file you want to process.

Claude does not edit video files directly; the MCP acts as the bridge that allows the assistant to call Kapwing’s editing tools and chain them into a project workflow.

How to connect Claude and install the skills (practical steps)

There are three common connection patterns depending on your Claude client:

  • Claude desktop app — Open Settings > Customize > Connectors, choose + Add, name the connector and paste the MCP server URL: https://www.kapwing.com/mcp. Continue and sign in when prompted.
  • Claude Code (terminal) — Run: claude mcp add --transport http kapwing https://www.kapwing.com/mcp. Then enter /mcp inside Claude Code, select the kapwing connector and authenticate.
  • Codex — Add and authenticate with: codex mcp add kapwing --url https://www.kapwing.com/mcp and codex mcp login kapwing. Complete browser auth promptly because requests can expire.

Once the connector is active, install the three skills. Each skill encapsulates a repeatable command sequence—upload, transcribe, translate and export—so you don’t have to re-specify every editing step for each file.

Operational checklist and best practices

  • Start with a short test clip. Run a 20–60 second sample through the flow to confirm transcript accuracy, speaker separation and voice-quality before batching larger jobs.
  • Review transcripts and speaker attribution. Automated transcription can misassign speakers in multi-person recordings; fix speaker labels and timing before generating translated captions or dubbed audio.
  • Decide how to handle original audio. Subtitle translation preserves the original voice; dubbing replaces it with generated speech. Pick the approach that matches brand tone and audience expectations.
  • Batch with installed skills for consistency. Using /captions, /translate and /dubbing produces consistent settings across videos and languages.
  • Include a human review step. Validate lip sync, untranslated terms, branded names and timing changes introduced by translation before publishing. Expect manual tweaks for multi-speaker or noisy audio.
  • Document authentication and token refresh. For production runs, track how connectors authenticate and refresh tokens so batch jobs don’t fail mid-process.

What to watch next

Monitor Kapwing MCP updates and new voice or language options before you scale. When integrating this into a content pipeline, codify your test criteria (transcript accuracy, speaker separation, acceptable lip-sync deviation) and log exceptions so your localization team can iterate quickly.

If you’re evaluating automation for large libraries, build a short pilot that includes noisy and multi-speaker examples. That will show where the automated flow saves time and where it needs human intervention.