Creators without footage
Explainers, listicles, product ads and faceless channels where the video starts from an idea or a link, not a recording.
ClipTalk makes short videos from inside your AI assistant. Connect its hosted MCP server once, then describe the video you want in Claude, ChatGPT, Cursor or any MCP client: it writes the script, generates visuals and voiceover, adds captions and music, and hands back an MP4. It's a creation tool first; clipping existing footage is possible but not the point.
Most tools on this site start with a long video and cut it down. ClipTalk starts with a sentence. You tell your assistant what the video is about, who should present it, what tone the voice should have and what music fits; ClipTalk's MCP server does the rest and returns a finished, captioned MP4 for TikTok, Reels or Shorts.
It can also take footage: import_media pulls a public link or social post into your library, and edit_video trims, cuts silences and changes captions. But there's no "find the best moments in this two-hour podcast" tool. For that, see best AI clipping agents.
estimate_cost and plan_from_template show the script and credit price without generating anything.Renders usually take a few minutes. Failed generations are refunded.
Add the server URL to your client. The first tool call opens a ClipTalk sign-in page; press Allow. OAuth 2.1 with PKCE, so there's no API key to paste or leak. New accounts start with free credits.
"A 20-second TikTok ad from this product page, upbeat voice, bold captions." The assistant picks the tools: list_templates to find a format, import_media for the link, list_voices for the narrator.
plan_from_template returns the full script and credit cost as a dry run. Nothing is created or charged until you say so, which is the step that keeps an agent from spending your balance on a guess.
create_from_template or create_video spends the credits and starts the render. get_video reports status and the download link when it's done, usually within a few minutes.
make_contact_sheet gives the assistant one image of frames across the video so it can check the result. Ask for changes and it calls edit_video (captions, music, voice, scenes, trims, silences) then render_video.
Grouped as the server documents them. "Spends" means the call uses credits; everything else is free to call.
list_templatesVideo formats and what each needsplan_from_templateDry run: script and price, no chargecreate_from_templateFinished video from a format + topicspendscreate_videoScene-by-scene build with your media or AI assetsspendsget_motion_dialectReference for custom motion-graphics sceneslist_videosYour videosget_videoStatus and download linkget_video_assetsTimeline, scenes, assetsmake_contact_sheetOne image of frames for checkingedit_videoCaptions, music, voice, scenes, layout, speed, trims, silencesspendsrender_videoRe-render to a new MP4spendsgenerate_mediaImage, clip, voiceover or talking presenterspendsimport_mediaFrom a public link or social postupload_mediaUpload a file to your librarylist_media / get_mediaLibrary, with transcriptslist_voicesSearch voices by language, accent, tonelist_musicBackground music by moodlist_charactersYour characters and the public castcreate_characterSave a reusable presenterspendsget_creditsBalance and planestimate_costPrice any request before creatingThere is no delete tool, and chat attachments aren't passed to MCP servers: share a link and use import_media. Private or age-restricted posts can't be imported.
It's a hosted server, so there's nothing to install. Point your client at the URL and sign in on first use.
claude mcp add --transport http cliptalk https://www.cliptalk.pro/mcp~/.cursor/mcp.json (snippet opposite)..vscode/mcp.json with "type": "http".codex mcp add cliptalk --url https://www.cliptalk.pro/mcpgemini extensions install https://github.com/payamsaremi/cliptalk-mcpAlso works in ChatGPT (Developer mode), Windsurf, Grok and any client that supports remote Streamable HTTP servers.
{
"mcpServers": {
"cliptalk": {
"url": "https://www.cliptalk.pro/mcp"
}
}
}Explainers, listicles, product ads and faceless channels where the video starts from an idea or a link, not a recording.
One of the few video tools with a documented remote MCP server, OAuth and cost-estimate tools, so an agent can make videos safely in a loop. See clipping MCP servers.
Save a character once and have every video fronted by the same face and voice, generated from the week's talking points.
Not for cutting the podcast, but good for intros, hook cards and recap videos around the clips you post. Track what it costs per video against the earnings calculator.
No moment detection or virality scoring. For turning a two-hour source into shorts, use OpusClip or ClipForge.
Tell the agent to always run estimate_cost first and to stop above a credit limit. There's no delete tool, so renders stay in the library.
Connecting is free and new accounts get starter credits. Creating, editing, rendering and generating media spend credits; plans and packs are on cliptalk.pro/pricing, and the amounts aren't published on the MCP page, so check there.
Not in the sense of finding highlights. It can import a public video and trim it with edit_video, but there's no moment detection. Pair it with a clipper if that's the job.
No. Chat attachments aren't passed to MCP servers. Share a public link and the assistant will call import_media, or use upload_media from a client that supports file paths.
Claude (claude.ai, Desktop, Claude Code), ChatGPT in Developer mode, Cursor, VS Code with Copilot, Windsurf, Codex CLI, Gemini CLI, Grok, and any client that speaks remote MCP over Streamable HTTP.
Instruct it to call estimate_cost or plan_from_template before any creating tool, and to ask you before spending more than a set number. Failed generations are refunded automatically.