15 min read

Runway MCP Server: The GitHub Projects, Explained

Runway has an official MCP server on GitHub and a hosted one. What each repo wraps, what it returns, what it costs, and which community forks still matter.

Runway MCP server projects explained, an AITuber guideAI VIDEO TOOLSRunway MCP,The GitHub ProjectsOfficial repo, hosted server, community forks. What each returns.AITuber
AITuber’s map of the Runway MCP projects: the official repo, the two hosted servers, the community forks, and what each one hands back.

Runway has an official MCP server, and it lives on GitHub. The repo is runwayml/runway-api-mcp-server, MIT licensed, written in TypeScript, and last committed on 10 August 2026. It wraps Runway’s developer API and gives Claude tools to generate video, images, and audio, upscale a clip, and edit one with Aleph. Runway also runs two hosted servers so you never have to clone anything: https://mcp.runwayml.com/mcp for people on a Runway plan, and https://dev.runwayml.com/mcp for developers on the API. We read all of it on 25 September 2026. Every route returns the same kind of thing, a clip of 2 to 10 seconds, or up to 30 with Seedance 2.5, plus images. None of them writes a script, narrates, captions, or posts. The community forks on GitHub add a tool or two on top and mostly lag behind.

This guide in 10 secondsTHE SHORT VERSIONThis guide in 10 secondsTHE ANSWEROfficial, open source, on GitHub.Plus two hosted servers from Runway itself.All of them return clips and images. Nothing more.THE FACTSWhat one call returnsA 2 to 10 second clip. Up to 30 on Seedance 2.5What it costsGen-4.5 is 12 credits a second. A credit is 1 cent on the APIFor a finished videoA pipeline server, such as AITuber, does the other five steps
The whole post in one card: what Runway ships, what it returns, and what it charges.

Key takeaways

QuestionShort answer
Is there an official Runway MCP server on GitHub?Yes. runwayml/runway-api-mcp-server, MIT, TypeScript, last commit 10 August 2026
Do I have to clone it?No. Runway hosts mcp.runwayml.com/mcp for plan users and dev.runwayml.com/mcp for API developers
What does it return?Video clips of 2 to 10 seconds, up to 30 on Seedance 2.5, plus images, upscales, Aleph edits, and text-to-speech
What does it not do?No script, no narrated voice-over across scenes, no captions, no publishing
Which community fork is worth a look?@mindstone/mcp-server-runway is maintained and installs with npx. The most-starred fork still targets a model Runway removed
What if I want a finished video from Claude?Connect the MCP server for AI video generation from AITuber. It writes, voices, captions, renders, and posts

What we checked, and what we found

We looked at Runway’s GitHub organisation, its developer docs, the runway.com MCP pages and help centre, npm and PyPI, and the MCP directories (PulseMCP, Glama, Smithery, mcp.so), all on 25 September 2026. Star counts and commit dates come from the GitHub API that day.

The picture is clearer than the search results suggest. Runway maintains one open-source server, one plugin, one skills repo, and two hosted endpoints. The community repos are older wrappers of the same API, a handful of forks of the official repo, and a couple of proxies for resellers. None of the directories lists all of this in one place, which is why “runway mcp server github” is a search at all.

Runway homepage showing Gen-4 video generation tools
The Runway homepage: Gen-4 class video generation and editing tools. Three MCP routes sit behind it.

The official repo: runwayml/runway-api-mcp-server

This is the one the GitHub search is looking for. It is a stdio and HTTP server that wraps the developer API, so it needs an API key in RUNWAYML_API_SECRET and bills developer credits, not your Runway plan.

The tools it exposes: runway_listModels, runway_generateVideo for image-to-video, runway_generateImage, runway_upscaleVideo, runway_editVideo for Aleph, runway_generateAudio for text-to-speech, runway_getTask, runway_cancelTask, and runway_getOrg. The manifest also lists runway_refreshTaskUrl. There is no text-to-video tool and no character performance tool in the repo, which surprised us. Its recommended defaults are gemini_image3_pro for images, seedance2 for video, and aleph2 for edits.

Installing means cloning and building. There is no @runwayml MCP package on npm, so npx does not work. The README recommends loading it into Claude Desktop as an unpacked extension, with a claude_desktop_config.json entry as the fallback, and it has an HTTP mode for hosting on Railway or behind Cloudflare Tunnel. It notes that ngrok is blocked by Anthropic’s connector backend.

Three caveats from the README are worth knowing before you build on it. Output links expire after 24 hours. Payloads are validated locally against the API’s OpenAPI enums, so an outdated clone rejects new models until you pull. And it polls inside the tool call to dodge Claude Desktop’s roughly four-minute tool timeout, which is the same problem every video server has to solve.

Two sibling repos round out the official set. runwayml/runway-mcp-plugin wraps the hosted server as a Cursor and Grok plugin with OAuth. runwayml/skills, 70 stars, is the skills-format alternative: npx skills add runwayml/skills, an API key, and a $10 minimum credit purchase.

The two hosted servers

If you are not building anything, skip the repo.

The hosted Runway MCP at https://mcp.runwayml.com/mcp launched on 27 May 2026. You add it as a custom connector in Claude or ChatGPT, or through the Cursor plugin, and sign in with your Runway account. Generations bill your plan credits and land in your Runway library, which is a real advantage over a chat log. It speaks Streamable HTTP only, and Runway’s help article says Explore Mode, the unlimited relaxed-rate mode on higher plans, does not apply over MCP.

The Dev MCP at https://dev.runwayml.com/mcp is for API builders. It authenticates with OAuth, bills developer credits, and exposes account tools such as whoami alongside generation. Runway’s docs give the one-line install for Claude Code:

claude mcp add --transport http --scope user runway-dev-mcp https://dev.runwayml.com/mcp

The community repos, and which ones still matter

ProjectStarsLast pushWhat it is
vericontext/vibeframe170Not checkedMulti-provider CLI and MCP. Runway is one backend among several
wheattoast11/mcp-video-gen1717 Jun 2025Text-to-video and image-to-video for Runway or Luma. Defaults to gen3a_turbo, which Runway removed on 30 July 2026
mindstone/mcp-servers1213 Sep 2026@mindstone/mcp-server-runway on npm, 23 tools, installs with npx -y. FSL-1.1-MIT licence
runway-mcp-server on PyPIn/a11 Nov 2025uvx runway-mcp-server, 12 tools including Aleph edits and 4K upscale. Its GitHub repo returns 404
@imbsq/runway-mcp on npmn/a23 Mar 2026A fork of the official repo that adds runway_textToVideo and an async mode. Its repo also returns 404
runapi-ai/runway-mcp031 Jul 2026Wraps RunAPI’s reseller proxy, not Runway directly

Stars and dates come from the GitHub API and the npm and PyPI registries on 25 September 2026. Star counts are absent where the project only exists as a package.

The most-starred single-purpose fork, wheattoast11/mcp-video-gen, is a good example of why forks age badly. Its default model is gen3a_turbo, and Runway removed that model, along with gen4_aleph, on 30 July 2026. It will fail until someone edits the model list. mindstone’s server is the one we’d point a developer at if they want npx and a broad tool set. Two packages on the list have no reachable source at all, which is a reason on its own to prefer the official repo.

What comes back, and what it costs

Every Runway route returns a clip or an image. The developer API prices per second of output, and a credit is one cent. From Runway’s pricing and model pages on 25 September 2026:

ModelCredits a secondClip lengthNotes
Gen-4.5122 to 10 sText-to-video and image-to-video, 720p, no audio parameter
Gen-4 Turbo52 to 10 sImage-to-video only, 720p
Aleph 228, minimum 56Video-to-videoEdits an existing clip. Launched 2 June 2026
Seedance 236 to 150Short36 at 480p and 720p, 40 at 1080p, 150 at 4K. Has an audio parameter
Seedance 2.520 to 68Up to 30 sPlus reference fees, 80-credit minimum, 1080p since 15 August 2026
Veo 3.140 with audio, 20 without4, 6, or 8 sUp to 1920 x 1080
Act-Two5Character performanceNot exposed by the official MCP repo

So 60 seconds of Gen-4.5 footage is 720 credits, or $7.20 in clips, before you write, voice, caption, or stitch any of it. Our best AI video APIs comparison prices the same minute across eight APIs.

On the hosted server you spend plan credits instead. Runway’s pricing page listed Free with 125 one-time credits, Standard at $15 a month or $12 billed yearly with 625 credits, Pro at $35 or $28 with 2,250, and Max at $95 or $76 with 9,500. Gen-4.5 on a plan is 60 credits per 5 seconds, so the Standard plan covers about ten of them.

30 seconds of Runway footage, by routeGen-4.5, developer API$3.60Gen-4 Turbo, developer API$1.50Gen-4.5, Standard plan360 of 625 credits30 SECONDS OF FOOTAGE, READ 25 SEPTEMBER 2026 (LOWER IS BETTER)Six 5-second clips. The plan bar is in credits, not dollars, so it is not to the same scale.All three routes stop at footage. Script, narration, captions, stitching, and the upload are still yours.For comparison, a 30-second narrated, captioned video with AI images on AITuber is about 35 credits.
Thirty seconds of Runway footage, priced by route. Good footage, and the start of the job rather than the end of it.

How to get a finished video from Claude instead

If the goal is a video on your channel rather than a folder of clips, connect a pipeline server. There is nothing to clone. AITuber’s server is hosted at https://mcp.aituber.app.

  1. Open Settings, then Connectors, in Claude Desktop or on claude.ai.
  2. Click Add custom connector and paste the server URL. The Claude MCP server for video page has the exact steps.
  3. Sign in with your AITuber account. No API key goes into Claude.
  4. Ask for a video. Say the topic, the length, the aspect ratio, and where to post it.
  5. Approve the tool call. AITuber writes the script, picks a voice from more than 1,300, generates the visuals, adds word-synced captions, renders the MP4, and publishes to YouTube, TikTok, Instagram, Facebook, Threads, or X.

For developers, the same account has a REST API that returns the finished MP4 from one call, and the MCP vs API guide covers when to use which. In Claude Code, claude mcp add --transport http aituber https://mcp.aituber.app connects the server.

Both connectors can live in the same client. Ask Runway for a Gen-4.5 opener when a scene wants it, and ask AITuber for the narrated video around it.

What a full pipeline returns

Finished videos made with AITuber

Every one of these was generated from a single prompt. No clip folder, no timeline, no export button.

Official repo vs hosted Runway MCP vs forks vs AITuber MCP

Official repoHosted Runway MCPCommunity forksAITuber MCP
Who runs itRunway, on your machineRunwayVolunteersAITuber
Sign-inAPI keyRunway accountAPI key or reseller keyAITuber account
InstallClone and buildOne URLnpx, uvx, or cloneOne URL
ReturnsClips, images, upscales, Aleph edits, TTSClips and imagesClips, sometimes text-to-videoA finished MP4 with voice and captions
Text-to-videoNot in the repoYesSome forksYes, up to 20 minutes
Script and captionsNoNoNoYes
PublishingNoNoNoYouTube, TikTok, Instagram, Facebook, Threads, X
BillingDeveloper credits, 1 cent eachPlan creditsDeveloper creditsAITuber credits
Price of entry$10 minimum purchaseFree, 125 credits, then $15 a monthPay as you go100 free credits, then $39 a month

The honest read. Runway’s official work is the best-built set of model servers we have looked at, and the hosted one is the easiest. The forks are mostly history now. And the whole set, official or not, ends at footage. For video from a Claude chat, AITuber is the route that finishes the job.

Who should use what

Use the official repo if you are a developer who wants to read the code, run it on your own infrastructure, or fork it. Pin your clone to the model list you tested, because the local validation rejects models it does not know.

Use the hosted Runway MCP if you pay for a Runway plan and want Gen-4.5 or Seedance from Claude or ChatGPT, with the results in your library.

Use the Dev MCP if you are building on the API and want account and generation tools in Claude Code without writing a tool layer.

Use a fork only for a tool the official repo lacks, such as text-to-video in a stdio server, and check the model list before you trust it.

Use AITuber if you want to ask Claude for a video and find it on your channel. The same MCP server for AI video generation works in ChatGPT, Cursor, and Codex, and our best MCP servers for AI video generation roundup compares it with Runway and the rest.

Frequently Asked Questions

Basics

Is there an official Runway MCP server on GitHub?

Yes. It is runwayml/runway-api-mcp-server, an MIT-licensed TypeScript server that wraps Runway’s developer API. Runway also hosts two servers, mcp.runwayml.com/mcp for plan users and dev.runwayml.com/mcp for API developers, so cloning is optional.

What does the Runway MCP server return?

Clips and images. Video clips run 2 to 10 seconds on Gen-4.5 and Gen-4 Turbo, 4 to 8 seconds on Veo 3.1, and up to 30 seconds on Seedance 2.5. The repo also exposes upscaling, Aleph edits, and text-to-speech. It does not write scripts, add captions, or publish.

Can Claude generate Runway videos?

Yes, through any of the three official routes. Claude calls the generate tool, polls the task, and returns a link to the clip. The link expires after 24 hours on the repo route, and on the hosted route the clip is saved to your Runway library.

GitHub and setup

How do I install the official Runway MCP repo?

Clone it, run npm run build, and set RUNWAYML_API_SECRET. Runway recommends loading it into Claude Desktop as an unpacked extension. There is no npm package, so npx does not work for the official server. For the hosted server, add https://mcp.runwayml.com/mcp as a custom connector instead.

Why does my Runway MCP fork fail on the model name?

Runway removed gen3a_turbo and gen4_aleph on 30 July 2026. Several forks, including the most-starred one, still default to gen3a_turbo. Change the model to gen4_turbo, gen4.5, or seedance2 and pull the latest OpenAPI enums.

Does the Runway MCP work in Claude Code?

Yes. The Dev MCP installs with one command, claude mcp add --transport http --scope user runway-dev-mcp https://dev.runwayml.com/mcp. The hosted server can be added the same way with its own URL.

Cost and comparisons

How much does a Runway clip cost through MCP?

On the developer API, a credit is one cent. Gen-4.5 is 12 credits a second, so a 10-second clip is $1.20. Gen-4 Turbo is 5 credits a second. On a plan, Gen-4.5 is 60 credits per 5 seconds against the 625 credits in the $15 Standard plan.

Runway MCP or AITuber MCP for a faceless channel?

AITuber, because a channel needs finished videos and Runway returns footage. Keep Runway connected for the occasional cinematic shot. Our MCP servers roundup ranks both.

Can I reach Runway models through fal.ai or Replicate instead?

Replicate hosts Runway’s Gen-4.5, Gen-4 Turbo, and Aleph models under the runwayml account, so Replicate’s MCP reaches them. fal.ai does not list Runway models. Runway’s own servers are the direct route.

Final verdict

The GitHub project people are searching for is real, official, and well made, and it has two hosted siblings that need no clone at all. Runway has done the developer work. What it built is a model server: one step of six, and a good one. If you make videos from footage, connect it and enjoy Gen-4.5 from a chat. If you want to ask Claude for a video and see it on your channel, connect AITuber’s Claude MCP server for video, and let Runway be the connector you reach for when a scene needs a cinematic shot.


How we researched this: we read Runway’s GitHub repositories, developer docs, pricing and model pages, MCP announcement, and help centre article on 25 September 2026. Star counts, licences, and commit dates come from the GitHub API that day, and package details from the npm and PyPI registries. Model removals are from Runway’s API changelog. The market figure is from Grand View Research. AITuber is a product we build, so every comparison is flagged, and the Runway details are kept accurate and current. If Runway changes its servers, this page will be updated.