The Reelrail MCP server

Reelrail speaks the Model Context Protocol, so an assistant can browse video models, price a shot, start a generation and hand you back a playable clip without leaving the chat. Jobs run against your workspace, under the same plan limits, credit ceilings and job history as the web app - they show up in your jobs list either way.

Server URL

https://api.reelrail.app/mcp

Before you connect

A Starter plan or higher. MCP access is off on the Free plan; a Free-plan token is refused with an upgrade message. See pricing.
Credits in the workspace. Generation tools spend the same credits the composer does; read-only tools are free.
Nothing is anonymous. Every request to /mcp carries either an OAuth access token or an organization API key. An unauthenticated call gets a 401 with the discovery pointer your client needs.

Tools

Everything below is available today: video, speech and captions, storyboards and share links - the same surface the web app has. Image tools are not listed because nothing renders an image yet; they arrive with image generation. Prefer plain HTTPS? The /v1 API reference is at /docs/api.

Tool	Scope	What it does	Cost
list_video_models	reelrail:read	Every video model Reelrail currently routes to, with resolutions, durations, audio and reference support, plus a price hint.	Free
estimate_video	reelrail:read	Price a generation before running it: credits, the upper bound Reelrail will hold, and your balance.	Free
generate_video	reelrail:generate:video	Start a generation with a prompt, duration, resolution, optional references and first/last frames. Pass pipeline edit, extend or upscale with parentJobId to work from a clip you already made. Returns a job id immediately.	Spends creditsSpends credits. A hold is placed when the job starts and settled against the provider's real cost when it finishes. An edit or an upscale is priced on the source clip.
get_job_status	reelrail:read	Poll a job, optionally waiting up to 60 seconds. Returns a playable link and a poster once the job succeeds.	Free
list_jobs	reelrail:read	Page through the workspace's jobs, filtered by status or model.	Free
cancel_job	reelrail:generate:video	Stop a queued or running job.	Free. A cancelled job is never charged and its hold is released.
upload_reference	reelrail:generate:video	Fetch a public image or video URL into the workspace so a later generate_video call can reference it.	Free. Counts against the workspace's storage retention window.
get_credit_balance	reelrail:read	Current balance, plan, per-job ceiling and monthly ceiling.	Free
synthesize_speech	reelrail:generate:video	Read a script aloud in one of twelve voices and store it as an audio asset a talking-head generation can use.	Spends creditsSpends credits: $0.015 per 1 000 characters at your plan's markup, never less than 1 credit. Repeat a requestId to retry without paying twice.
transcribe_captions	reelrail:generate:video	Transcribe an asset or a finished job's clip to WebVTT and store the captions as a downloadable asset.	Spends creditsSpends credits: billed per audio minute, which is 1 credit for any realistic clip.
attest_asset_rights	reelrail:generate:video	Record that the person in an uploaded photo may be used. A talking-head job is refused until its avatar image carries this.	Free. Only call it after the person using the assistant has confirmed the likeness statement themselves.
create_sequence	reelrail:generate:video	Create a storyboard: one model, shared references and defaults, and up to 24 shots that each render as their own job.	Free. Nothing renders until run_sequence.
get_sequence	reelrail:read	Read a sequence with every shot, its job status and a playable link per finished shot.	Free
list_sequences	reelrail:read	Page through the workspace's sequences, newest first.	Free
run_sequence	reelrail:generate:video	Submit the sequence's unrendered shots, as many as the plan's concurrency cap allows; the rest come back with the reason they were skipped.	Spends creditsSpends credits: one job per submitted shot, at the same price generate_video would charge.
regenerate_shot	reelrail:generate:video	Re-shoot one shot, optionally with a new prompt, preset or parameter overrides.	Spends creditsSpends credits: a fresh job for that shot. Repeat a requestId to retry without paying twice.
archive_sequence	reelrail:generate:video	Put a sequence away. It stops appearing in lists and can no longer be run.	Free. The rendered jobs and clips are kept.
create_share_link	reelrail:generate:video	Publish a finished job or sequence at a public URL that anyone with the link can watch - prompt, model, parameters and the clip, never your reference files or workspace details.	Free. Optional expiry from 1 hour to 90 days; revocable at any time.
list_share_links	reelrail:read	What the workspace has published, with each link's URL and view count.	Free
revoke_share_link	reelrail:generate:video	Stop a share link working. The public page 404s from the next request on.	Free

What you are approving

The first time a client connects, Reelrail asks you to approve its request on https://app.reelrail.app/consent. You can approve a subset, and you can take it back at any time from Connected apps.

Scope	Means
reelrail:read	List available models, read your jobs and their results, and see your credit balance.
reelrail:generate:video	Start video generations, upload references and cancel jobs. Spends credits from the workspace below.
offline_access	Keep working after you close the browser, without asking you to sign in again. Revoke it any time under Connected apps.

Connect your client

Claude.ai and Claude Desktop (custom connector)

Settings -> Connectors -> Add custom connector.
Paste the server URL and click Connect.
Sign in to Reelrail in the browser window, approve the request, and you are done.

Server URL

https://api.reelrail.app/mcp

Claude Desktop (JSON config, via mcp-remote)

Open the Developer settings and edit claude_desktop_config.json.
Add the entry below, then restart Claude Desktop. A browser window opens for the OAuth flow.

claude_desktop_config.json

{
  "mcpServers": {
    "reelrail": {
      "command": "npx",
      "args": ["mcp-remote", "https://api.reelrail.app/mcp"]
    }
  }
}

Claude Code

Add the server, then run /mcp inside Claude Code and authenticate in the browser.
Prefer a headless setup (CI, a shared box)? Skip OAuth and pass an organization API key instead.

OAuth

claude mcp add --transport http reelrail https://api.reelrail.app/mcp

Organization API key

claude mcp add --transport http reelrail https://api.reelrail.app/mcp \
  --header "Authorization: Bearer rr_org_xxxxxxxxxxxxxxxx"

ChatGPT

Settings -> Connectors (developer mode) -> Create.
Enter the server URL, choose OAuth authentication, then Connect and approve.

Server URL

https://api.reelrail.app/mcp

Cursor and Windsurf

Add the server to the editor's MCP config and reload. Both open the OAuth flow.

Cursor (.cursor/mcp.json)

{
  "mcpServers": {
    "reelrail": { "url": "https://api.reelrail.app/mcp" }
  }
}

Windsurf (mcp_config.json)

{
  "mcpServers": {
    "reelrail": { "serverUrl": "https://api.reelrail.app/mcp" }
  }
}

MCP Inspector (for testing)

Run the inspector, paste the server URL, set the transport to Streamable HTTP.
Open Auth settings and run Quick OAuth Flow, then call tools/list.

Terminal

npx @modelcontextprotocol/inspector

Without OAuth: organization API keys

Scripts, CI jobs and any client that cannot run a browser flow can send an organization key instead. Create one under API keys - it starts with rr_org_, belongs to the workspace rather than to you, and carries whatever permissions you gave it. Treat it like a password: it is shown once and it spends real credits.

JSON-RPC tools/list over curl

curl -sS https://api.reelrail.app/mcp \
  -H "Authorization: Bearer rr_org_xxxxxxxxxxxxxxxx" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Revoking the key stops every client using it on the next request. OAuth grants are revoked separately under Connected apps.