The Best Video MCP Servers for AI Agents in 2026: 18 Compared

An MCP server is what lets Claude, ChatGPT or Cursor use another app as a tool. For video, the server you connect decides what your agent can actually hand back: a five-second clip, a talking avatar, a template filled with data, or a finished video that is already scheduled on TikTok.

We compared 18 official video MCP servers on October 22, 2026. For each one we read the vendor's own documentation, sent an MCP handshake to the remote endpoint to check it is live and how it signs in, and noted what comes out, whether it can research or post, and what plan it needs. Then we ran one real job end to end through our own server and timed it.

Disclosure: we make Revid. It is first on this list because it is the only server here that researches what goes viral, makes the finished video and posts it, all in one connection. The closest alternative, OpusClip's Agent Opus, also makes and posts finished videos, and we cover it second. For every other kind of video job we say which server we would pick instead, and why.

#How we compared them

A server made the list if it is official (published by the company that runs the video product), remote (you connect with a URL, nothing to install), and documented. We left out the hundreds of community wrappers around one model's API: they need your own API keys, often go stale, and an agent cannot sign in to them with one click.

For each server we checked:

• What it delivers: a raw clip, an avatar video, a template render, an edit of your footage, or a finished video with script, voice, captions and music.
• Sign-in: OAuth (your agent opens a sign-in page, no key to paste) or an API key in a header. OAuth is what Claude.ai, ChatGPT and most hosted clients expect.
• Research: can the agent look up what already performs before it makes anything?
• Publishing: can the agent post the result to social accounts, or only hand you a file?
• Plan: what you need to pay before the agent can render.

Facts come from each vendor's docs and pricing pages, read on October 22, 2026, and from an unauthenticated handshake to each endpoint the same day. We did not create accounts on competing products, so the only hands-on test below is our own. Prices change often: treat them as a snapshot and check the vendor's page.

#The comparison table

Two patterns stand out. Almost every video product now ships an official remote MCP server with OAuth, so "does it have MCP" no longer separates them. And most of them cannot post: the agent makes a file, and a person still has to download it and upload it to each app. In this table, three publish natively: Revid, Agent Opus and OpusClip (which only clips a long video you already have). Only Revid also researches what to make.

#1. Revid: research, make and post finished videos

Best for: founders, marketers and agencies who want the agent to deliver a video that is ready to go out, and to post it.

Revid's MCP server is the same engine as the Revid app, exposed as 30 tools. A typical agent run uses four of them: search_viral_videos to see what already works on the topic, render_video to make the video, get_project_status to wait for it, and schedule_publish or publish_now to post it. In between, calculate_credits prices the job before anything is spent.

What sets it apart:

• It starts from what works. search_viral_videos searches viral posts on YouTube, Instagram, TikTok, X, Reddit, Threads and LinkedIn and returns each post's transcript, hook, engagement and style. Searches cost no credits. No other server in this comparison has a viral video search.
• It delivers a finished video. render_video takes 14 documented workflows: a prompt, a script, an article or any link, an audio file, a song, an avatar, a product ad, captions for your own video, clips from a long video, motion transfer and more. The output has the voiceover (1,000+ voices in 70+ languages, or your cloned voice), captions in 30 styles, generated music that fits the script, and AI footage from models such as Seedance 2.5, MiniMax Hailuo H3, Wan 3.0 and Google Veo 3.
• It posts. list_social_accounts finds the connected TikTok, YouTube, Instagram and X accounts, schedule_publish posts at an exact time up to a year ahead or in the next free slot of your queue, and get_publish_status confirms the post went live with its link.
• It keeps characters consistent across a series. create_episode makes a whole episode of a series with the same characters, places and props in every scene, and the next episode from the same series later.
• One-click sign-in everywhere. OAuth 2.1 with PKCE and dynamic client registration, so Claude, ChatGPT, Claude Code, Cursor and VS Code connect with your Revid login and no API key. Other clients send an API key in a header. The same capabilities are in the REST API and the revid-cli npm package.

videos made through the Revid API, MCP and CLI

39,000+

October 2025 to October 2026

videos rendered on Revid every month

100,000+

videos published to social media from Revid every month

30,000+

MCP tools: research, create, manage, publish

30

Accounts

50100150

148

Oct 25NovDecJan 26FebMarAprMayJunJulAugSep

Source: Revid analytics (project_built events from the API, MCP and CLI), checked October 22, 2026

Show the data

	Accounts
Oct 25	52
Nov	42
Dec	41
Jan 26	58
Feb	75
Mar	92
Apr	110
May	114
Jun	124
Jul	135
Aug	134
Sep	148

#Our test: one sentence to a finished video

We gave Claude one brief through the Revid MCP server on October 22, 2026, and timed each step.

The brief: a 30-second vertical explainer for founders on what an MCP server is, made with motion graphics, voiceover, captions and generated music.

What we would fix: in two moments the captions sit on top of the graphics. One more render_video call with captions moved, or a minute in the Revid editor, would solve it. We stopped before posting to keep our accounts clean; the next call would have been schedule_publish with a scheduledTime. Motion graphics are the slowest kind of render on Revid: a video made of AI footage or stock visuals usually takes a few minutes.

Where Revid is not the best pick. If you need a presenter on camera reading a long script in many languages, HeyGen is more specialized. If you want one shot from one specific model with every parameter of that model exposed, a model platform such as Runway or fal.ai gives you more control per call. And Revid needs a paid plan to render (from $39 a month, MCP included); search and price estimates work without one.

Connect Revid to your agent · API docs

#2. Agent Opus by OpusClip: finished videos, posted

Best for: turning an idea or a script into a ready-to-post video when you do not need research, and posting it to many networks.

OpusClip runs a second server, Agent Opus (api.opus.pro/api/agent-mcp, OAuth, 29 tools), separate from its clipping server. Its page describes turning "any idea, topic, or script into a finished, ready-to-post video" with stock or cloned voices, 25+ visual style presets and brand assets, then publishing now or on a schedule to YouTube, TikTok, Instagram, Facebook, LinkedIn, X and Dailymotion. MCP access needs the AgentOpus Pro plan or higher and uses the same credits as the app. Compared with Revid, its documentation shows no viral research tool, and it posts to more networks (Facebook, LinkedIn and Dailymotion as well).

#3. HeyGen: avatar and presenter videos

Best for: a person on screen, from a stock avatar or your own digital twin, reading a script; translating a video into other languages.

HeyGen's official server at mcp.heygen.com/mcp/v1/ signs in with OAuth and lists 97 tools in its docs: avatar videos, avatar creation, voice cloning and design, templates, video translation and lip sync, AI clipping, and a Video Agent (create_video_agent) that builds script, scenes, B-roll, captions and music from one prompt. That makes it the closest to a finished-video server after Revid. It works on every HeyGen plan with premium credits. Two limits to know: HeyGen's own developer notes call MCP sign-in "trial-scale" and point to its CLI with an API key for volume, and no tool posts to social media.

#4. Synthesia: presenters, and a video agent

Best for: training, onboarding and corporate explainers with an AI presenter, and teams already on Synthesia.

Synthesia's server (mcp.synthesia.io/mcp, public beta, OAuth per user) creates avatar video drafts from a script, manages templates and brand kits, and runs Syren, a beta agent that makes a finished video with script, visuals, motion graphics, music and an optional presenter, up to 3 minutes at 720p. It needs the Starter plan ($29 a month, $18 billed yearly) or above, and Synthesia is one of the video apps listed in ChatGPT's directory. It does not post to social media.

#5. Pictory: finished videos from text

Best for: turning a blog post or script into a stock-footage video with voice and captions, in landscape, vertical or square.

Pictory's hosted server launched on September 9, 2026, with OAuth and 14 tools such as create_video_from_text, preview_video_from_text, search_media and list_voices. Videos run 8 to 500 seconds, and renders take 5 to 15 minutes, more with AI clips. It cannot post to social accounts. Its docs disagree on billing (the API subscription from $79 a month billed yearly, or an eligible plan or trial), so check before you build on it.

#6. InVideo: an agent in a timeline editor

Best for: agents that should edit, color and mix inside a timeline and hand off a Premiere Pro, Final Cut Pro or DaVinci Resolve project.

InVideo's current server at mcp.invideo.io/mcp (OAuth) puts the agent inside invideo's cloud editor. Separately, invideo's ChatGPT plugin turns a prompt into a finished video, which is why it is the name ChatGPT users hear most. Neither posts to social media, and invideo does not say which plan includes MCP.

#7. OpusClip: clips from long videos, with posting

Best for: podcasts, webinars and streams you want cut into vertical clips.

OpusClip's server (mcp.opus.pro/mcp, OAuth) submits a long video, lists the clips its AI picked, applies edits and exports, and it can schedule posts to YouTube Shorts, TikTok, Instagram and more. The connector needs the Pro plan ($29 a month) or Business. It starts from footage you already have: it does not make a video from an idea or a script.

#8. Raw clip generators: Runway, Higgsfield, fal.ai, Pika, Hedra, LTX

Best for: one shot, or a few, from a specific model, when your agent or a person assembles the final edit.

• Runway (mcp.runwayml.com/mcp, launched May 27, 2026) uses your Runway plan credits and reaches Gen-4.5, Gen-4 Turbo, Veo 3.1, Seedance and Kling, plus product ad and UGC presets. A separate Runway Dev MCP helps developers research models and debug tasks.
• Higgsfield (mcp.higgsfield.ai/mcp) covers 30+ image and video models, Soul characters and workflow presets, always at standard credit rates.
• fal.ai (mcp.fal.ai/mcp-relay) searches and runs any of its 1,000+ models, billed per run. It cannot read local files, and resubmitting a job bills it again.
• Pika (mcp.pika.art/mcp) makes clips and multi-shot videos up to 30 seconds and edits them on a timeline.
• Hedra (mcp.hedra.com) generates video, images, speech, music and sound effects from its catalog, including avatars with lip sync. It needs a subscription.
• Kling also answers at kling.ai/mcp with OAuth, but Kling has not published docs or a tool list for it yet.
• LTX Studio (app.ltx.io/mcp) generates and edits clips: reframe, extend, retake, relight, remove objects.

What they have in common: each call returns a clip of a few seconds, with no script, voiceover, captions or music around it, and nothing posts it.

#9. Template rendering: Shotstack and Creatomate

Best for: the same branded layout rendered hundreds of times with different data (real estate listings, product feeds, sports scores).

• Shotstack (mcp.shotstack.io, in beta, OAuth or API key) has 15 tools: render from a JSON timeline, create and render templates, open an editor inside the chat, and generate AI clips, images, speech and music. Your agent can assemble a finished video, but it does the assembly. Pay as you go starts at $10 for 50 render minutes; plans start at $39 a month.
• Creatomate (api.creatomate.com/mcp, OAuth or a project API key) has 8 tools to list, create, edit and render templates. Plans start at $54 a month for 2,000 credits.

You design the template first; the agent fills it. JSON2Video and Plainly offer the same kind of rendering through local MCP servers with an API key.

#10. Avatars from a script: VEED Fabric

Best for: a quick talking-head clip from one of 24+ stock characters and a voice.

VEED's "Fabric" server (veed.io/api/v1/mcp, OAuth) has 7 tools, from list_characters to create_fabric_video, and returns an MP4 link. It costs about 8 credits per second of video. It makes the avatar clip only.

#11. Editing your own footage: Descript and Kapwing

Best for: footage you recorded, which needs captions, cleanup and cuts.

• Descript (api.descript.com/v2/mcp, in the Claude directory since May 2026) edits through its AI editor: filler words, retakes, Studio Sound, captions, translation, and timeline export to editors. It shares to a Descript link rather than to social apps. Paid plans start at $16 a month billed yearly.
• Kapwing (kapwing.com/mcp, all users since October 12, 2026) uploads, captions, translates subtitles, dubs and exports MP4, and works on the free plan with limits (720p, watermark, 4-minute cap).

#Also checked

• Finished-video makers without publishing: Zebracat (mcp.zebracat.ai/mcp, 42 tools), VideoGen (mcp.videogen.io/mcp, 49 tools) and Lumen5 (lumen5.com/mcp) make full videos from a script or text; none of their MCP servers posts to social accounts.
• Design tools: Canva's server (in the Claude directory) can autofill an existing video template and export MP4, but cannot generate a video. Adobe's connector resizes, cuts and stitches clips.
• Newer servers that make and post: Faceless.so, MakeAIVideo, AITuber and AutoShort also create short videos and publish them (AutoShort's own pages disagree on whether posting works yet). They are younger and smaller, and none documents a viral video search. Reap clips long videos and posts the clips.
• Not available: OpenAI's Sora API was removed on September 24, 2026, Clipchamp has no MCP or public API, and Luma's official MCP server has not been updated since April 2025.

#How to connect Revid in a minute

The server URL is the same everywhere: https://www.revid.ai/api/mcp.

1. Claude: Settings, then Connectors, then Add custom connector. Paste the URL, click Connect and sign in with your Revid account.
2. ChatGPT: in Settings, turn on developer mode under Advanced, create a connector with the URL and OAuth, then sign in.
3. Claude Code: run claude mcp add --transport http revid https://www.revid.ai/api/mcp, then /mcp and Authenticate.
4. Cursor and VS Code: use the one-click install buttons on revid.ai/mcp, then sign in.

Then ask in plain words, for example:

Questions people ask

What is the best MCP server for making videos?

For finished short-form videos that also get posted, Revid: one remote server researches what goes viral, makes the video with voice, captions, music and AI footage, and publishes or schedules it on TikTok, YouTube, Instagram and X. For avatar presenter videos pick HeyGen or Synthesia, for single cinematic clips Runway, Higgsfield or fal.ai, for template rendering Shotstack or Creatomate, and for editing your own footage Descript or Kapwing.

Can ChatGPT make videos?

ChatGPT writes the script; to deliver a finished video it needs a video tool connected. With Revid connected, ChatGPT makes the video from one sentence, with voice, captions and music, and can post it. In ChatGPT settings, turn on developer mode, add https://www.revid.ai/api/mcp as a connector and sign in with your Revid account.

How do I make a video with Claude?

Add Revid as a custom connector in Claude (Settings, Connectors, Add custom connector, https://www.revid.ai/api/mcp), sign in, and describe the video. Claude calls render_video, waits with get_project_status and returns the video link, and can post it with publish_now or schedule_publish.

Which video MCP servers can post to TikTok, Instagram and YouTube?

Of the 18 servers in our table (checked October 22, 2026), three publish natively: Revid (TikTok, YouTube, Instagram, X and webhooks, now or at a scheduled time), Agent Opus by OpusClip, and OpusClip's clipping server. Smaller servers such as Faceless.so, MakeAIVideo, AITuber and Reap also post. The rest hand back a file. Posting-only servers such as Blotato, Postiz or Upload-Post can sit next to a video server, but they do not make the video.

Do I need an API key to connect a video MCP server?

Not for most of them in 2026. Revid, HeyGen, Runway, Higgsfield, fal.ai, Descript, Kapwing, OpusClip and the other servers in our table sign in with OAuth: your agent opens a sign-in page and you approve it. API keys remain for clients that cannot do OAuth and for server-side code.

How much does it cost to make a video through MCP?

Every server here bills in its own credits. On Revid, the agent can call calculate_credits to price a job before it starts. Our 30-second motion graphics test cost 122 credits, about $2.40 at the Growth plan's monthly price. Plans start at $39 a month and include the MCP server, the API and the CLI.

Last verified October 22, 2026. Vendors ship MCP changes every week: if a fact here is out of date, write to [email protected] and we will correct it with a dated note.