Eddie is a programmatic video editor with a CLI, a headless SDK and an MCP server. Point an agent at raw footage — it cuts, captions, scores and exports to Premiere, Resolve or MP4. No timeline to scrub.
Three ways in
The same engine behind every surface. Script it from a shell, embed it in your backend, or hand it to an autonomous agent as a set of tools. Uploads go straight to S3 with a presigned URL — the key never holds cloud credentials.
Create, import, prompt and export in four commands. Progress on stderr, ids on stdout — pipe it into anything.
# capture just the id
id=$(eddie new \
--import ./footage/*.mov)
eddie prompt "$id" \
"make a 6min rough cut"
eddie export "$id" \
"Rough Cut" --format mp4The core is dependency-free — Node built-ins and fetch. Trade an sk_ key for a short-lived token, then drive projects server-side.
import { Eddie } from "@eddieai/cli" const eddie = new Eddie(key) const p = await eddie.newProject( { import: files }) await eddie.prompt(p.id, "cut the best 10 minutes")
Eddie ships a Model Context Protocol server. Connect Claude or any MCP client and every editing action becomes a typed tool.
// claude_desktop / any MCP host "eddie": { "url": "https://mcp.heyeddie.ai/api/mcp", "auth": "sk_live_…" }
Prefer no key? The hosted connector signs in with OAuth — set it up in Claude or ChatGPT.
Humans + AI, one project
Eddie is not a black box that spits out a file. Agent and human work the same project on a shared, link-native timeline — the agent does the heavy cutting, the editor steers, comments and finishes in their NLE.
the agent proposes · the human disposes · the project is the shared surface
The tool surface
Not "generate a video." Discrete, composable operations an agent can reason about and chain — each one does one editorial job, and reports what it actually did.
create_edit_resultappend_to_editcreate_video_sequencefind_transcript_segmentslayout_editadd_breakadd_captionsadd_music_trackadd_brollsset_multicam_anglesgenerate_voiceovercorrect_transcripttrim_segmentanalyze_editexport_editexport_selectsbuild_documentget_credit_balancelist_editsget_transcriptFrom prompt to render
A prompt run is server-driven. The client starts it and streams the activity as it lands — every step is a real editorial decision you can read back, refund-safe and idempotent.
Presigned S3 upload plus ffprobe. Sources analyse in parallel. Prompting unlocks only when analysis is complete.
The agent reads transcripts, picks soundbites and builds to a length it chose on purpose.
Captions, music, b-roll, angle switches, dubs. Each is a scoped call that leaves the rest of the cut untouched.
Render an MP4, or export interchange for Premiere, Resolve and Final Cut. Cuts relink to the originals.
What it can do today
Auto-syncs angles from shared audio and cuts to whoever is speaking. One interview, every camera.
Karaoke-accurate burned-in captions, plus translated dubs into 30+ languages that re-caption themselves.
Finds coverage across the project, or generates and places subclips against the words on screen.
FCPXML, Premiere .prproj, Resolve .drp, EDL. Generated server-side, cuts intact.
Best soundbites to vertical 9:16 with smart reframing, and 16:9 from the same edit.
Query the transcript like a chat: find moments, pull selects, build a shot log or a Word doc.
Mint a workspace key, install the CLI, and put an editor in your agent’s hands in five minutes.
npm i -g @eddieai/cli · eddie login · Settings → Integrations → API keys
Looking for named editing agents you add inside Eddie instead? Browse the agent marketplace.