The Complete Guide to Using ClipSpeedAI Inside Claude (MCP)

For years, turning a long video into short-form clips meant leaving whatever you were working in, opening a separate editor, scrubbing a timeline, burning captions, exporting nine or ten files, and then bouncing to yet another tool to write titles and schedule posts. Model Context Protocol collapses that. With the open clipspeed-mcp server connected to Claude, you can say "clip this Twitch VOD into vertical shorts, then write me TikTok hooks for the best three" — and it happens inside one conversation, start to finish.

This is the definitive, first-hand reference for doing exactly that. ClipSpeedAI publishes an open MCP server — clipspeed-mcp on npm — that connects to Claude and exposes ten video tools the assistant can call on your behalf. To be precise about what this is: it is a nominative integration built on the open protocol. It is not an official Anthropic product, plugin, or partnership, and nothing here is endorsed or sponsored by Anthropic. It is simply a server that speaks MCP, the same protocol Claude — and every other MCP-capable assistant — understands. That distinction matters, and it is also the good news: because the integration lives on an open standard, the workflow in this guide works with Claude today and survives whatever model or client you use next.

You are the hero of this workflow, not the AI. ClipSpeedAI does the video labor and Claude does the language labor, but the direction — what to clip, which moments matter, what the copy should say, when to post — stays with you. This guide walks the entire path: what MCP actually is, why clipping from inside your assistant changes the math, how to get a key, how to install on every major client, a copyable prompt library, the ten tools in full, best practices, and a troubleshooting section that covers every real error you can hit.

What MCP is, in plain English

Model Context Protocol (MCP) is an open standard that lets an AI assistant talk to outside tools and services in a consistent way. Instead of every app inventing its own bespoke plugin format, a service publishes an MCP server that describes the tools it offers — each with a name, a description, and a set of inputs. An MCP-capable assistant (the host, such as Claude Desktop or Claude Code) reads that description and can then call those tools when a task calls for them.

The mental model most people find useful: MCP is to AI assistants roughly what a USB-C port is to a laptop. The port is a shared shape; anything built to that shape plugs in and just works. You do not rewire the laptop for each new device, and the device maker does not build a different plug for every laptop. MCP is that shared shape for AI tools. Once ClipSpeedAI's server is plugged into Claude, Claude "sees" ten new capabilities and knows when and how to use them — you never have to memorize an API, write a curl command, or copy-paste JSON. You describe what you want in plain language, and Claude picks the right tool with the right inputs.

There are three moving parts worth naming, because they show up in the setup steps and the errors:

• The host is the app you talk to — Claude Desktop, Claude Code, Cursor, Cline, or Windsurf. It runs the model and manages the conversation.
• The client is the piece inside the host that speaks MCP and manages the connection to a server. You rarely think about it directly; it is what "adds a server" for you.
• The server is the tool provider — here, clipspeed-mcp. It advertises the ten tools and executes them by calling ClipSpeedAI's API with your key.

The important thing for a creator is that MCP is not "an API you have to code against." A traditional API is a contract for programmers: you read docs, write requests, handle responses. MCP flips that. The server describes itself to the model, and the model does the calling. You stay in natural language the entire time. If you want the deeper background before diving in, read What Is Model Context Protocol (MCP)? A Creator's Guide. For the rest of this guide, that one-minute version is all you need.

Here is the whole picture in one line — how a plain-English request travels from Claude to a finished clip on your machine:

Why creators should care

You could always open the ClipSpeedAI web app, paste a URL, and get clips. So why do it from inside Claude at all? Because clipping is rarely the whole job — it is the first step of a chain, and every other step is language work that Claude is already good at. The outcome, not the technology, is the reason.

Here is the difference in concrete terms. In the web app, the pipeline is: paste URL → get clips → leave → write titles somewhere → leave → schedule somewhere. Every arrow is a context switch, and context switches are where a Sunday of batching quietly turns into a Sunday lost. With the MCP server, the entire chain is one conversation:

• One context, no tab-hopping. The transcript, the clip scores, and the copy all live in the same thread. Claude already read the transcript to help you, so its titles and hooks are grounded in what was actually said — not generic guesses that could apply to any video.
• Claude does the writing layer. ClipSpeedAI returns scored, captioned 9:16 clips. Claude then writes the hooks, titles, descriptions, hashtags, carousel captions, tweets, and even a newsletter — all referencing the real clip content, in your voice if you tell it your voice.
• It closes the loop. Through the same tools, Claude can submit the job, poll until clips are ready, download the finished files to your machine, and schedule them to five platforms. You approve; it executes. You are never copy-pasting a URL between four apps.
• It is scriptable and repeatable. "Every Monday, find the fastest-growing gaming VOD from this week, clip it, and draft five TikTok captions" is a single instruction, not a Monday-morning ritual you have to remember to run.

There is also a compounding effect that is easy to miss on day one. When the clip pipeline and the copywriting live in the same thread, Claude accumulates context about your channel — your niche, your tone, which hooks you liked last week, which platforms you actually post to. That context makes each subsequent session sharper. For a full worked example of the chain end to end, see Claude + ClipSpeedAI: Turn One Livestream Into 30 Days of Content.

Getting your API key

Everything downstream needs an API key, so start here. Keys are gated behind a paying plan — this is a deliberate anti-abuse measure, and it is cheap to clear. There is no free-tier key by design.

You need a paid plan or the $1 trial

The free 30-minute plan lets you use the web app but does not issue an API key. To get one, you need either an active paid plan or the $1 three-day trial, which unlocks Full Pro for three days. The trial is the fastest path for testing the MCP integration end to end — a dollar buys you enough runway to clip a few VODs, run the full discover→clip→download→schedule chain, and decide whether it fits your workflow.

1. Go to clipspeed.ai and start the $1 trial (or sign in if you already have a paid plan).
2. Open your dashboard and navigate to the API key page: clipspeed.ai/dashboard/api-key.
3. Click to generate a key. It will look like csai_live_xxxxxxxxxxxx.
4. Copy it immediately and store it somewhere safe — treat it like a password. Anyone with the key can spend your credits.

The csai_live_ prefix is how ClipSpeedAI and the MCP server recognize a production key at a glance. When you paste it into a config or environment variable later, keep the whole string intact — the prefix and the random tail together are the credential.

Full developer documentation, including the raw HTTP endpoints if you ever want to bypass the MCP layer entirely and call the API yourself, lives at clipspeed.ai/developers. You do not need it for anything in this guide, but it is there when you want to script something custom.

Installation for every client

The same npm package, clipspeed-mcp@1.0.0, powers every client. Pick your setup below. In all cases the key idea is identical: register the server and give it your csai_live_ key through an environment variable or config field. The tool surface — all ten tools — is the same no matter which client you use.

Claude Desktop (easiest — guided installer)

Claude Desktop has the smoothest path. The package ships an installer that writes itself into your Claude Desktop config and prompts for your key:

npx clipspeed-mcp@1.0.0 install

Follow the prompt, paste your csai_live_ key when asked, then fully quit and reopen Claude Desktop. You will see ClipSpeedAI's ten tools available in the tools menu. That is the entire setup — no JSON editing by hand.

If you would rather edit the config yourself, Claude Desktop stores its MCP configuration in claude_desktop_config.json. The location depends on your OS:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

You can find and open it inside the app under Settings → Developer → Edit Config. The manual block is identical to the editor JSON shown further down.

Claude Code (terminal-native)

If you live in the terminal, Claude Code registers the server with one command and reads your key from the environment:

claude mcp add clipspeed -- npx -y clipspeed-mcp@1.0.0
export CLIPSPEED_API_KEY=csai_live_xxxxxxxxxxxx

Add the export line to your shell profile (~/.zshrc or ~/.bashrc) so it persists across sessions. Start a new Claude Code session and the tools are live. To confirm the server registered, run claude mcp list — you should see clipspeed in the output.

Cursor (JSON config)

Cursor uses a JSON config block. Add this to ~/.cursor/mcp.json (create the file if it does not exist), or use the per-project .cursor/mcp.json to scope the server to a single project:

{
  "mcpServers": {
    "clipspeed": {
      "command": "npx",
      "args": ["-y", "clipspeed-mcp@1.0.0"],
      "env": {
        "CLIPSPEED_API_KEY": "csai_live_xxxxxxxxxxxx"
      }
    }
  }
}

Save, then open Cursor's MCP settings panel (Settings → MCP) and confirm clipspeed shows a green/connected indicator. Reload if it does not appear immediately.

Cline (VS Code extension)

Cline runs as a VS Code extension and keeps its MCP servers in its own settings file. Open the Cline panel, click the MCP Servers icon, and choose to edit the configuration. Cline uses the same mcpServers shape:

{
  "mcpServers": {
    "clipspeed": {
      "command": "npx",
      "args": ["-y", "clipspeed-mcp@1.0.0"],
      "env": {
        "CLIPSPEED_API_KEY": "csai_live_xxxxxxxxxxxx"
      }
    }
  }
}

The config file lives under your VS Code global storage for the Cline extension; the in-app "Edit MCP Settings" button opens it for you so you do not have to hunt for the path. Save and Cline reconnects.

Windsurf (JSON config)

Windsurf (Codeium's editor) exposes MCP under Settings → Cascade → MCP, and writes to ~/.codeium/windsurf/mcp_config.json. Add the same server block:

{
  "mcpServers": {
    "clipspeed": {
      "command": "npx",
      "args": ["-y", "clipspeed-mcp@1.0.0"],
      "env": {
        "CLIPSPEED_API_KEY": "csai_live_xxxxxxxxxxxx"
      }
    }
  }
}

Click "Refresh" in the Windsurf MCP panel after saving, and the ClipSpeedAI tools appear alongside your other servers.

Hosted transport (no local install)

If you do not want to run the package locally at all — for example on a machine without Node, or in a hosted agent environment — connect over HTTP to the managed endpoint. Nothing installs on your machine; Claude talks to the hosted server directly:

claude mcp add --transport http clipspeed https://api.clipspeed.ai/mcp \
  --header "Authorization: Bearer csai_live_xxxxxxxxxxxx"

The hosted transport exposes the identical ten tools. It is the leanest option when you want zero local footprint, and it is the right choice on headless boxes, CI environments, or any host where installing Node is a hassle. Editors that support a remote or streamable-HTTP MCP transport can point at the same URL with an Authorization: Bearer header.

Verifying it worked

Do not assume the connection is live just because the install command exited cleanly. Two quick checks confirm it end to end.

1. Confirm the tools are registered. In Claude Desktop, open the tools menu and look for the ClipSpeedAI tools. In Claude Code, run claude mcp list and confirm clipspeed appears with a connected status. In Cursor, Cline, or Windsurf, open the MCP panel and look for a green/connected indicator next to clipspeed. If the server is listed but shows an error, it usually means the key is missing or malformed — jump to Troubleshooting.

2. Confirm the key works with a live call. The cleanest smoke test is a read-only tool that touches your account. Ask Claude:

Check my ClipSpeedAI usage — how many credits do I have left this month?

Claude calls get_usage and reports your plan and remaining credits. If that comes back with real numbers, your key is valid, the server is wired up correctly, and you are ready to clip. If it errors with an authentication message, the key is not reaching the server — again, see Troubleshooting. Verifying with get_usage is deliberate: it is read-only, so it cannot cost you credits or create anything, and it fails loudly if auth is wrong.

Your first prompts

Once the server is connected, you talk to Claude in plain language. You do not name the tools; Claude infers which to call from what you ask. A short sequence of first prompts to feel the whole loop:

Make your first clips:

Clip this YouTube video into vertical shorts with karaoke captions:
https://www.youtube.com/watch?v=XXXXXXXX

Claude calls clip_video with the URL and your caption preference. It returns a job that runs the full pipeline: score every moment, cut the best, transcribe with Whisper, caption, and reframe to 9:16.

Poll, then download:

Are those clips done yet? If so, download the top three to my Downloads folder.

Claude calls get_clip_status (or list_clips) and, once finished, download_clip for the highest-scored clips. The files land in ~/Downloads.

Then let Claude write around the clips:

For the top clip, write me 5 TikTok hooks, a YouTube Shorts title,
and 8 hashtags. Base them on what's actually said in the clip.

This is where the "one conversation" payoff shows up: Claude already has the transcript context, so the copy is grounded, not generic. That single detail — "base them on what's actually said" — is worth internalizing, because it is the whole reason to clip inside the assistant instead of beside it.

The prompt library

The prompts below are real and copyable. They map to the five-stage flow — discover, clip, download, schedule, restyle — and then to the multi-step chains that combine them. Paste one, swap in your URL or niche, and go. For a much larger set of copywriting prompts to run on the clips once they exist, see 45 Best Claude Prompts for Viral Shorts.

Discover — find something worth clipping

Find the fastest-growing Valorant video from the last week and give me the
link plus a one-line reason it's blowing up.

What's the top trending gaming stream this week that I could clip? Just
surface the single best candidate, not a list.

Clip — turn a URL into shorts

Clip this Twitch VOD into vertical shorts with hormozi captions:
https://www.twitch.tv/videos/XXXXXXXX

Clip this YouTube video, aim for the 6 highest-energy moments, karaoke
captions, 9:16. Then list them ranked by score:
https://www.youtube.com/watch?v=XXXXXXXX

Take this Kick stream and pull only the funniest moments as square 1:1
clips with beasty captions: <URL>

Download — get the files locally

Download the top three clips from that last job to my Downloads folder.

Only download clips that scored above 80. Skip the rest.

Schedule — queue posts across platforms

Schedule the top clip to TikTok tomorrow at 6pm and to YouTube tomorrow
at 8pm.

Spread these four clips across the next four days: one per day to TikTok
and Instagram, starting Monday at 5pm.

Show me everything I currently have scheduled this week, then cancel the
Thursday LinkedIn post.

Restyle — test a new caption look

That top clip did well. Re-render its captions in fire style so I can A/B
test it, then download the new version.

Redo clip #2's captions in youshaei style — I want a cleaner, more
editorial look for LinkedIn.

Multi-step workflows — the whole chain in one instruction

These are where the integration earns its keep. Each is a single message that fans out into a sequence of tool calls plus writing.

Find the fastest-growing Minecraft video from the last week, clip it into
vertical shorts with beasty captions, and once they're done, download the
two highest-scored clips to my Downloads folder.

Clip my latest Twitch VOD (URL below) into shorts with hormozi captions.
When they're ready, list them ranked by score. For the top four, write a
TikTok caption, a YouTube Shorts title, and an Instagram caption each.
Then schedule one clip per day to TikTok, YouTube, and Instagram starting
tomorrow at 5pm.
<URL>

Clip this podcast episode, pull the 5 most quotable moments as 9:16 clips
with cinematic captions, and for each one draft a tweet and a LinkedIn
post. Download all five, but don't schedule anything yet — I'll review.
<URL>

Here's my stream from last night. Clip it, then build me a full week of
content: rank the clips, write platform-native copy for the top seven,
and schedule them one per day across TikTok and Instagram at 6pm. Show me
the queue when you're done.
<URL>

An end-to-end workflow, step by step

To make the chain concrete, here is a single Sunday session walked through beat by beat. Assume you streamed for three hours on Twitch and want a week of content out of it. The whole session is just five tool calls Claude makes for you:

1. You paste the VOD and set the brief. "Clip my Twitch VOD from last night into 9:16 shorts with hormozi captions. Aim for the highest-energy and funniest moments." Claude calls clip_video with the URL, the aspect ratio, and the caption style. Behind the scenes, ClipSpeedAI ingests the three hours, scores every moment, cuts the best candidates, transcribes them with Whisper, burns the captions, and reframes to vertical.
2. You wait — briefly — and Claude polls for you. A three-hour source is real work, so the job takes time. You say "let me know when they're ready," and Claude calls get_clip_status until the job reports done. You do not sit refreshing a page.
3. Claude lists the clips, ranked. Once finished, Claude calls list_clips and shows you the clips with their viral scores. Now you can see, at a glance, which three or four moments the system rated highest.
4. You direct the edit. "Download the top five to my Downloads folder, and for clips 1 and 3 redo the captions in beasty style — those are the loud gaming moments." Claude calls download_clip five times and restyle_captions on the two you flagged, then downloads the restyled versions.
5. Claude writes the copy. "For each of the top five, write a TikTok caption, a YouTube Shorts title, and three hashtags — grounded in what's actually said, in my dry no-emoji voice." Because Claude has the transcript in context, the copy references the real moment, not a template.
6. You schedule the week. "Schedule one clip per day to TikTok and Instagram starting tomorrow at 6pm." Claude calls schedule_post for each clip and platform. (This step needs each platform connected once via OAuth in your ClipSpeedAI dashboard — see Troubleshooting if it errors.)
7. You review and adjust. "Show me the full queue." Claude calls list_scheduled_posts. You spot one you want to move and say "cancel the Wednesday TikTok post" — Claude calls cancel_scheduled_post. Done.

Start to finish, that is one conversation. You never opened a video editor, never scrubbed a timeline, never copy-pasted a URL across four apps. You made every creative decision; the tools executed them. That is the shape of the whole product, and every workflow in the prompt library is a variation on it.

The 10 tools reference

Here is every tool the clipspeed-mcp server exposes, what it does, and when Claude reaches for it. You never call these directly — you describe the outcome and Claude selects them — but knowing the surface helps you write sharper prompts and understand what an error is actually telling you.

Caption styles and aspect ratios

When you ask for a caption look, Claude passes one of six styles to clip_video or restyle_captions:

• karaoke — word-by-word highlight as spoken; broadly safe default.
• hormozi — bold, high-contrast, keyword-highlighted; strong for business and education.
• beasty — punchy, high-energy style tuned for retention on entertainment and gaming.
• fire — animated, attention-grabbing emphasis.
• youshaei — clean, minimal, creator-editorial look.
• cinematic — understated, film-style captions.

Aspect ratios are 9:16 (vertical, the default for shorts), 1:1 (square), and 16:9 (landscape). Just say what you want — "make these square for the feed" — and Claude sets it. Transcription runs on Whisper, and clip storage auto-deletes over time, so download anything you want to keep.

Best practices

The integration rewards a few habits. None are complicated; all of them save credits, time, or headaches.

Which caption styles convert

Honestly, the best style depends on your audience, and anyone who tells you there is a single winner is guessing. That said, some patterns hold up: hormozi tends to perform for business, finance, and education content because the keyword highlighting reinforces the point being made. Beasty and fire suit high-energy entertainment and gaming, where the motion matches the pace. Karaoke is the reliable all-rounder when you are not sure. Youshaei reads as clean and editorial, which plays well on LinkedIn and for talking-head clips. The right move is not to theorize — it is to A/B test. Clip once, then use restyle_captions to render the same moment two ways and post both. Let the numbers decide.

Batch to save context switches, not credits

Credits are billed by source minute, so batching does not save credits — a three-hour VOD costs the same whether you clip it once or think about it all week. What batching saves is your time. Feed a long source once, get all the clips, and let Claude write the copy for all of them in the same thread while the transcript is fresh in context. One session of "clip, rank, write for the top seven, schedule the week" beats seven separate sessions.

Manage credits deliberately

Because one source minute equals one credit, a long VOD can eat a plan fast. Two habits help: check get_usage before submitting a big source so you are not surprised, and prefer restyle_captions over re-clipping when you only want a different caption look — restyling re-renders an existing clip and does not re-bill the source minutes. If you are on the $1 trial and testing, clip a shorter source first to confirm the whole loop before spending on a three-hour stream.

Hosted transport vs. the npm package

The local npm package and the hosted HTTP transport expose the identical ten tools, so choose on operational grounds, not features. Use the npm package when you are on a normal workstation with Node and want everything running locally. Use the hosted transport when you are on a headless server, a machine without Node, or a hosted agent environment, or when you simply want zero local footprint. Neither is "better"; they are the same product with different plumbing.

Keep your key secure

Your csai_live_ key is a spendable credential. A few rules keep it safe:

• Never paste it into a chat message — put it in the config or environment variable, where the client reads it, not into the conversation with the model.
• Never commit it to Git or check a config file containing it into a public repo. If you use a per-project .cursor/mcp.json, add it to .gitignore.
• Never include it in a screenshot you share — the dashboard masks the key for this reason.
• Rotate on exposure. If a key ever leaks, revoke and regenerate it at clipspeed.ai/dashboard/api-key. Anyone holding it can spend your credits.

Troubleshooting every error

Here is every real error you can hit, what it means, and the exact fix. Work top to bottom; the earlier ones are the most common on first setup.

"No API key" or authentication errors

If Claude reports it cannot authenticate, the server is not receiving a valid key. Check, in order:

• The key is present. In Claude Code, confirm echo $CLIPSPEED_API_KEY prints your csai_live_ key in the same shell you launched Claude from. In Cursor, Cline, or Windsurf, confirm the env block in your JSON has the key.
• The key is complete. A truncated paste (missing the tail) fails auth. Regenerate at clipspeed.ai/dashboard/api-key and paste the whole thing, prefix included.
• You restarted the host. Environment and config changes only take effect on a fresh session — fully quit and reopen Claude Desktop, or start a new Claude Code session.

"Paid plan required" / subscription required — key generation blocked

If the dashboard will not issue a key, you are on the free 30-minute plan, which does not include API access. Start the $1 three-day trial (Full Pro) or upgrade to Starter/Pro, then return to the API key page. The gate is intentional; there is no free-tier key. If you already pay and still see this, confirm you are signed into the correct account — the plan is tied to the account, not the browser.

Scheduling won't post to a platform

Posting requires a one-time OAuth connection per platform. schedule_post will accept and queue the post, but it can only go live once the platform (TikTok, YouTube, Instagram, X, or LinkedIn) is linked to your ClipSpeedAI account. Connect each platform once in your ClipSpeedAI dashboard, then schedule. If you queued a post before the platform was connected, it shows up as failed in list_scheduled_posts with the reason — connect the platform and reschedule. Clipping and downloading do not require any platform connection — only posting does — so if your goal is just to get files locally, you can skip this entirely.

Out of credits or an expired plan

Two different things can stop a clip. If your paid plan or $1 trial has lapsed, the API returns a subscription_required error (HTTP 403) with an upgrade link — restart the trial or upgrade, then retry. If your plan is active but you are low on minutes, run get_usage first to see the balance; one source minute = one credit, so a long VOD consumes a lot. A job your remaining credits can't cover won't finish — get_clip_status will report it as failed. Fixes: wait for renewal, upgrade a tier, or clip a shorter source.

A clip job was accepted but then failed

A clearly malformed link is rejected immediately with a validation_error. But a well-formed URL pointing to a video ClipSpeedAI cannot actually read — private, unlisted-with-restrictions, region-locked, age-gated, deleted, or behind a login wall — is accepted, queued, and then fails during processing. Poll get_clip_status: a failed status with a message tells you what went wrong. Confirm the link opens in a normal, logged-out browser and points to a supported source (YouTube, Twitch, Kick, or a direct upload). If it is a live stream that has not ended, wait for the VOD.

Rate limits and long jobs

Two things people mistake for errors:

• A long video still "processing." A three-hour VOD legitimately takes time to score, cut, transcribe, and caption. If get_clip_status says processing, it is working — just poll again in a bit. Ask Claude to "check again in a few minutes."
• Rate limits on rapid-fire calls. If you fire a burst of tool calls (for example, "download all ten right now"), you can hit a request rate limit. Space the calls out — download the top few you actually need rather than the whole batch at once, and let a paused job finish before hammering status checks.

"Package not found" on install

If npx clipspeed-mcp@1.0.0 install errors with a package-not-found message, the usual culprits are a typo in the package name, a stale npm cache, or no network. Confirm the name is exactly clipspeed-mcp, that you have Node and npm installed (node -v and npm -v should both print versions), and that you can reach the registry. Clearing the cache with npm cache clean --force and retrying resolves most transient cases. If you cannot install Node at all, skip the package entirely and use the hosted HTTP transport instead.

Two-factor prompt when publishing

Some platforms require a two-factor confirmation before a post actually goes out. If a scheduled post stalls waiting on verification, that is the platform's security step, not a ClipSpeedAI failure. Complete the 2FA/verification in your ClipSpeedAI dashboard for that platform connection, and the queued posts proceed. Reconnecting the platform (a fresh OAuth pass) also clears a stale or expired authorization that can cause the same stall.

If you hit something not covered here, the raw API behavior, error codes, and limits are documented at clipspeed.ai/developers.

Frequently asked questions

What is the ClipSpeedAI MCP server?

ClipSpeedAI publishes an open Model Context Protocol server called clipspeed-mcp on npm (v1.0.0). Once you connect it to an MCP-capable assistant like Claude, the assistant gains 10 tools it can call on your behalf — turning a long video URL into scored, captioned, 9:16 clips, checking their status, downloading them, discovering trending videos, and scheduling posts. It is a nominative integration, not an official Anthropic product or partnership: it is simply an open MCP server that speaks the same protocol Claude understands.

Is this an official Anthropic or Claude integration?

No. ClipSpeedAI is an independent product and clipspeed-mcp is an open MCP server it publishes. There is no official partnership, endorsement, or sponsorship with Anthropic. We use the name Claude only to describe the real assistant the server connects to. Because the server implements the open Model Context Protocol, it also works with any other MCP-capable assistant, not only Claude.

Do I need a paid plan to use ClipSpeedAI inside Claude?

Yes. Generating an API key requires an active paid plan or the $1 three-day trial (which unlocks Full Pro). The free 30-minute plan lets you use the web app but does not issue an API key. You can start the $1 trial at clipspeed.ai, then create a csai_live_ key at clipspeed.ai/dashboard/api-key.

How do I install clipspeed-mcp in Claude Desktop?

Run npx clipspeed-mcp@1.0.0 install in your terminal. The installer writes the server into your Claude Desktop config and prompts for your csai_live_ API key. Restart Claude Desktop, and the 10 ClipSpeedAI tools appear. If you prefer Claude Code, run claude mcp add clipspeed -- npx -y clipspeed-mcp@1.0.0 and export CLIPSPEED_API_KEY.

Does the ClipSpeedAI MCP integration only work with Claude?

No. Because clipspeed-mcp implements the open Model Context Protocol, it works with any MCP-capable assistant — Claude Desktop, Claude Code, Cursor, Cline, Windsurf, and other MCP hosts. The setup steps differ slightly per client, but the same 10 tools are exposed everywhere. Building on the open protocol means your workflow survives model and tool changes.

What can Claude do that ClipSpeedAI alone cannot?

ClipSpeedAI does the heavy video work — scoring every moment, cutting the best, transcribing with Whisper, captioning, and reframing to 9:16. Claude adds the language layer on top: after the clips come back, Claude can write hooks, titles, descriptions, hashtags, carousel captions, tweets, and newsletters, then (through the same tools) submit new jobs, download finished clips, and schedule posts. The clip pipeline and the copywriting happen in one conversation.

How does ClipSpeedAI pricing and credit usage work?

Pricing is $1 for a 3-day Full Pro trial, Starter at $15/month for 150 minutes, and Pro at $29/month for 350 minutes. One minute of source video equals one credit — there are no multipliers. The free plan gives 30 minutes per month for the web app. You can check your remaining credits and plan at any time by asking Claude to call the get_usage tool.

Which platforms can Claude schedule ClipSpeedAI clips to?

The schedule_post tool supports five platforms today: TikTok, YouTube, Instagram, X, and LinkedIn. You connect each platform once through your ClipSpeedAI account, then Claude can schedule finished clips, list scheduled posts, and cancel them via the list_scheduled_posts and cancel_scheduled_post tools.

Which caption styles convert best?

ClipSpeedAI offers six caption styles: karaoke, hormozi, beasty, fire, youshaei, and cinematic. Karaoke is the safe all-purpose default with word-by-word highlighting. Hormozi (bold, keyword-highlighted) tends to perform for business and education content; beasty and fire are high-energy styles tuned for retention on entertainment and gaming; youshaei is a clean editorial look; cinematic is understated for film-style clips. The honest answer is that the best style depends on your audience, so use restyle_captions to A/B test two looks on the same clip without re-clipping the source.

Is it safe to give an MCP server my API key?

Your csai_live_ key is stored locally — in your Claude Desktop config, your shell environment, or your editor's MCP settings file — and is sent only to ClipSpeedAI's own API. Treat it like a password: never paste it into a chat message, a shared repo, or a screenshot, and never commit it to Git. If a key is ever exposed, revoke and regenerate it at clipspeed.ai/dashboard/api-key. Anyone holding the key can spend your credits, so keep it out of anything public.

What do I do if a clip job fails?

A clearly malformed link is rejected right away with a validation_error. A well-formed URL that ClipSpeedAI cannot actually read — a private, region-locked, age-gated, deleted, or login-walled video — is accepted and queued, then fails during processing; poll get_clip_status and it reports a failed status with a message. Confirm the link opens in a normal, logged-out browser and points to a supported source (YouTube, Twitch, Kick, or a direct upload). If you are low on credits, run get_usage first — one source minute equals one credit, so a long VOD consumes a lot, and a job your credits can't cover won't finish. If your plan or $1 trial has lapsed you'll get a subscription_required error with an upgrade link.

Version history

Last updated: 10 July 2026. This is living documentation — it is versioned and kept current as the integration ships new tools, platforms, and fixes. Safe to bookmark and return to.

This page tracks the integration as it ships. When new tools, platforms, or caption styles land, this version history and the tool table above are where they are recorded — so pin the latest clipspeed-mcp version listed here.

Related resources & docs

This guide is the hub. These are the pages it connects to — the docs you need to set up, and the sibling articles that go deeper on each part of the workflow.

Official docs & the package

• ClipSpeedAI Developer Docs & API — the raw HTTP endpoints, error codes, and limits behind the MCP layer.
• Your API key dashboard — generate, view, and rotate your csai_live_ key.
• clipspeed-mcp on npm — the open MCP server package (v1.0.0).
• Open the ClipSpeedAI app — the web app, if you want to clip without the MCP layer.

The rest of the cluster

• What Is Model Context Protocol (MCP)? A Creator's Guide
• Claude + ClipSpeedAI: Turn One Livestream Into 30 Days of Content
• 45 Best Claude Prompts for Viral Shorts
• The Best MCP Servers for Creators in 2026
• Claude vs ChatGPT for Content Creators
• The AI Creator Stack (2026)