Connect Reelium to Claude

Reelium puts your clip review queue behind the Model Context Protocol, so Claude can work it with you: page the unreviewed clips, read what the picker found in each one, set a clip's frame and caption style, move the in and out points, render, and post to an account you have already connected. You sign in through your browser, so there is no key to copy into a config file.

Nineteen tools are registered on the server today, behind six scopes. The full reference — the rate limits, the closed design vocabulary, the starter prompt that ships with it — is on the MCP server page.

01 Claude and claude.ai

Reelium is a remote MCP server, which Claude calls a custom connector. In Claude's settings, add a custom connector and give it the endpoint above. Claude then sends you to a Reelium sign-in page, and after that to the consent screen described below. Approve it and the connector is live in your next conversation.

The exact menu wording, and which plans offer custom connectors, are Anthropic's to state and they change independently of Reelium — so rather than reprint them here, the current steps are in Anthropic's guide to custom connectors. Reelium's half of the exchange is the endpoint and the two discovery documents, and those are the same whichever Claude you use.

02 Claude Code

One command adds it:

claude mcp add --transport http reelium https://reelium.ai/api/mcp

Then run /mcp inside Claude Code and choose to authenticate. That opens the same browser flow as above — no key is handled at any point. Adding a server this way records it for the current project; pass --scope user instead to have it available in every project on your machine. Anthropic documents the command and its scopes.

There is a second way in, and it is the right one only for a headless or CI setup where no browser can open: a key you mint in Reelium under Settings, MCP access, sent as Authorization: Bearer …. Reach for it last. A key passed on a command line is stored in clear text in your Claude Code configuration file and lands in your shell history; the browser flow leaves nothing on disk for you to rotate. A key expires after 90 days by default, and a key created with no scopes ticked can read clips and nothing else.

What you approve

The consent screen is Reelium's, and it is worth reading rather than clicking through, because three things on it are easy to miss.

• Reelium verifies no application, including Claude. Anyone can register an app and choose the name and logo it shows, so the screen says so plainly. The one fact an app cannot choose is where your sign-in code is sent, and that address is printed in large type. Check it against the app you just used. For Claude Code that is a local address on your own machine; for claude.ai it is Anthropic's.
• You can untick any single permission. Every box arrives ticked, because granting what the app asked for is what you set out to do — but a grant is not all-or-nothing. Decline publishing and keep the rest, and the server issues the narrower grant. The two consequential ones are flagged where you cannot miss them: rendering and analysis are marked as costing money, and publishing as posting where you cannot take it back.
• The approval has no expiry date. It stays in force until you revoke it. The screen also names which Reelium account is granting, which matters if you keep more than one signed in.

These are the six scopes, in the words the consent screen itself uses — this page renders the same constant, so the two cannot drift apart:

Read clipsclips:read

Read your clips, analysis results, preview links and agent rules — including picks that did not enter the review queue. Changes nothing.

Create & review clipsclips:write

Create new draft clips in your account, mark clips reviewed, and delete rejects (recoverable for a short while). Does not render or publish.

Renderclips:render

Start renders and check their progress. Each render costs money.

Publishclips:publish

Post to YouTube / TikTok / Instagram.

Analyse VODsclips:analyze

Submit a stream for analysis. Costs Studio minutes; the longest source it accepts depends on your plan.

Frame & styleclips:design

Read frames from your source video and set a clip's crop, layout, captions look and trim; the camera you confirm is shared with everyone who clips that channel. Does not render or publish.

Before it can change anything

There are two gates, not one. A scope decides which tools Claude may call at all. An agent policy decides which of those calls finish without a question. Reelium enforces both on the server, so nothing Claude tells itself can widen either one.

That is the intended starting state rather than a fault — but read the next sentence before you try your first render, because it is the part that surprises people. With no policy stored, each of those four actions comes back as a question and the tool call does not complete. There is no argument that overrides it, and answering “yes, go ahead” in the conversation does not release it either: the gate reads the stored policy, so the way past it is to record one. These are the four, and the tools each covers:

Framing and style

update_clip_design, confirm_clip_layout

Trimming

trim_clip

Rendering

render_clip

Publishing

publish_clip

Settings, MCP access, Agent policy is where that answer is recorded: which of the four actions may complete unasked, which platforms a publish may reach, a minimum score, whether burnt-in captions are required, and the title words that hold a clip for a human decision. Anything the rules do not cover comes back to you as a question.

Only a signed-in person can author one. The endpoint that stores a policy refuses any request carrying an Authorization header outright — a key cannot mint a key, and it cannot author a policy either, so an agent has no route to raising its own ceiling.

Captions come first

Rendering refuses a clip that has no stored captions, and no tool in this set can create any — captions are synced in the Reelium dashboard, by you. So the order matters: sync the captions, then ask Claude to render. A clip Claude created for you is a draft with no captions yet, which means “find me some moments and render them” stops at the render until you have been through the dashboard.

An explicit render-without-captions flag exists to get past that refusal. It is there so that nobody arrives at a captionless clip by accident, not as the normal path — and if your policy requires captions to publish, a clip rendered that way is refused at the publish anyway.

Managing access

Everything lives on one page once you have an account: Settings, MCP access. Connected applications are listed there with a Disconnect button, which is what revokes the approval you gave Claude. Keys are minted and revoked there too, and the agent policy is recorded there. Revoking is the whole undo — a scope you never granted was never reachable, and a clip that has already been posted stays posted.