Model Context Protocol

Clipwave MCP for any AI agent.

Connect Clipwave to your agent and generate images and video — and drive the Media Studio editor — straight from your prompts. One hosted endpoint, your credits.

Claude CodeClaude DesktopCursorWindsurf+ any MCP client

Endpoint · Streamable HTTP, JSON-RPC 2.0

https://clipwave.io/api/mcp

Create an API key →REST API docs

Connect in three steps.

Create an API key

Grab a cw_live_ key from /dashboard/api-keys. (OAuth clients can skip this and just sign in.)

Add the connector

Point your agent at the Clipwave endpoint below and pass the key as a Bearer header.

Connect & create

Claude Code

claude mcp add clipwave \
  --transport http https://clipwave.io/api/mcp \
  --header "Authorization: Bearer cw_live_…"

Cursor / Windsurf / generic client

{
  "mcpServers": {
    "clipwave": {
      "url": "https://clipwave.io/api/mcp",
      "headers": { "Authorization": "Bearer cw_live_…" }
    }
  }
}

Just describe it. Clipwave builds it.

Generate images

"5 product shots of my serum, soft studio lighting, 1:1"

Generate video

"a 15-second skincare UGC ad with captions"

Edit in Media Studio

"add scene 2, crossfade between clips, then export"

Build a full ad

"script → scenes → voiceover → lip-sync → one MP4"

The toolset

Generation + the full Media Studio editor. The endpoint self-briefs on connect, so agents learn these automatically.

Media generation

list_models

Model catalog with capabilities. Call first — never guess a model_id.

media_forge_generate

Generate video or image. Async → returns a job_id.

job_status

Poll a job until completed, then read result.video / result.images.

Media Studio — projects

video_editor_create_project

Create a project (shared library + timelines).

video_editor_list_projects

List the user's projects.

video_editor_create_video

Add another video to a project.

video_editor_get_state

Inspect a video's timeline + library + context.

video_editor_set_context

Set the creative brief the agent reads each turn.

Media Studio — build

video_editor_instruct

One natural-language instruction (generate a scene, add subtitles, set a transition…).

video_editor_add_clip

Place a library asset on the timeline by id.

video_editor_add_reference

Import a public image/video/audio URL into the library.

video_editor_upload_file

Upload raw file bytes into the library.

video_editor_export

Stitch the timeline into one MP4. Returns a video_url.

A typical agent run

list_models                          → pick an exact model id
video_editor_create_project          → { project_id }
video_editor_instruct  "scene 1: …"  → generates + places a clip
video_editor_instruct  "scene 2: …, continue the look"
…add subtitles / set a transition
video_editor_get_state(video_id)     → verify the timeline
video_editor_export                  → { video_url }

Questions, answered.

Which agents are supported?

Anything that speaks MCP — Claude Code, Claude Desktop, Cursor, Windsurf, or your own client. Streamable HTTP, JSON-RPC 2.0.

Key or login?

Both. Headless/server agents use a cw_live_ Bearer key. Clients that support OAuth can just have the user sign in — discovery is automatic via /.well-known/oauth-protected-resource/api/mcp.

Whose credits does it use?

The connected account's. The agent generates on your Clipwave credits; failed jobs are refunded automatically.

What can it actually do?

Generate images and video, and fully drive Media Studio — create projects, add scenes, set transitions and captions, and export a finished MP4.

Does it self-explain?

Yes. On connect, the initialize response returns an instructions guide the agent reads automatically (auth, when to use each tool, the editor flow).