What it is: the Lucidream MCP server gives any MCP-capable agent — Claude, ChatGPT, OpenClaw or the one you built yourself — direct access to Lucid's production pipeline. Nine tools: ingest an episode from a URL, file, or RSS feed; transcribe, cut, add captions, reframe clips; and pull the creator's full context — brand template, formats, preferences — so what the agent produces is on-brand by default.
Why this matters: we tested the alternative before building this. We gave a fully-equipped, best-in-market agent 40 minutes and every tool it asked for, and told it to clip a podcast episode on its own. It burned roughly 1,000x the tokens our pipeline uses, tried four different ways to download the video, and delivered one clip — with the speaker's head cropped out of frame and no branding. Agents are extraordinary at deciding what should happen to content. They are the wrong tool for doing it. The MCP is the missing hand: your agent decides, Lucid produces — all from a single chat.
Built for the agents themselves: Lucid's MCP is built for your agent. Every feature we built, every tool like clipping and cropping, every error message is a self-healing instruction with the fix spelled out. The MCP details exactly how to use Lucid's tool stack, best practices to extract your podcast's best moments, and even how to write the perfect hook to go along with the post. All schema problems are reported in one pass, not discovered across three round-trips. Anything an agent can't do alone — connecting an account, payment — returns a URL to hand its human. Credit costs are stated on every tool, and your key carries daily caps and a kill switch, so a retry loop can never surprise you.
Get started: API key at signup, first successful tool call in under two minutes. Trial credits included.