Livepeer Agent
livepeer.org/agent
Connect your agent over MCP and reach image, video, audio, 3D, editing, and production tools across the Livepeer network.
Overview
Livepeer Agent is the network's own agent surface: one MCP server that gives any
agent harness the ability to generate, edit, and finish moving image and audio.
Point your agent's connector settings at
https://agent.livepeer.org/api/mcp/creative and the tools appear alongside
whatever else it already has. No key is needed: ask it for an activation code
by email to unlock the demo credit.
It is aimed at people who are already working in an agent — in an editor, a chat harness, or their own product — rather than at a separate web app to switch into. The network answers on the other side with models, compute, and a growing library of production skills, so the agent gains capabilities instead of the user gaining another tab.
What you can build
- Generation across media — image, video, audio, and 3D from a single connector, without wiring up a model provider per modality
- Editing and finishing — background removal, colour grading, overlays, subtitle burn-in, concatenation, and muxing, exposed as tools the agent can call in sequence
- Playbooks — production-ready multi-step workflows that run in the Agent Console rather than being reassembled from scratch each time
- Agent capabilities in your own product — point an app you already ship at the same server and it gains the same image and video workflows
Developer surface
- How it runs — a hosted MCP server; add the URL to any agent harness that speaks the protocol, no SDK to install
- What ships with it — hundreds of capabilities spanning generation, editing, rendering, CAD, and speech, all callable as MCP tools
- How to extend it — the same connector embeds the workflows in your own product, and a playbook library packages repeatable multi-step jobs
Powered by Livepeer
Every job routes to the open network of independent compute providers rather than a single vendor's cluster. That is what lets one connector cover this much surface area: capabilities are contributed and served by orchestrators across the network, so the tool list grows without the agent needing a new integration each time.