Deterministic, verified media workflows for software agents. Inspect → plan → execute → verify. Zero silent failures.
View on GitHub ↗Homepage ↗Deterministic media workflows for software agents. Inspect intent. Replay plans. Verify outcomes. Zero silent failures.
Agent Media turns goals such as "make this vertical, broadly compatible, and under 25 MB" into a versioned semantic plan. The plan can be inspected before execution, serialized across process boundaries, replayed through FFmpeg, and verified against fresh output metadata afterward.
FFmpeg is the first backend. It is deliberately not the agent-facing abstraction.
Prerequisites: Node.js 22+, ffmpeg and ffprobe on PATH.
Then ask for the outcome, not the flags — "make talk.mp4 vertical for shorts, under 25 MB" — and the agent gets a plan it can show you, an execution, and a verification that either passed or names what failed.
New here? Read the usage guide. It covers all three surfaces task by task.
Most media wrappers stop after a subprocess exits successfully. Agent Media treats that as only one step in a safer contract:
Property What the agent gets ---------------------- -------------------------------------------------------------------------------------------- Semantic planning Goals and reasons instead of raw FFmpeg flags Portable Media IR Validated JSON that can be reviewed, persisted, and replayed Outcome verification Structured checks for duration, geometry, size, audio, codec, and pixels Structured recovery Stable error codes and every failed verification check in one report Observable execution Monotonic progress through SDK callbacks, CLI NDJSON, and MCP notifications One implementation The SDK, JSON CLI, MCP server, and high-level workflow
From the project README.
Add the radar badge to your README — it shows your project was picked up by MCP Radar and links to this page:
[](https://mcp.liqiwa.com/s/HadiAlMarzooq--agent-media.html)
One local-first MCP server for agent working state — clipboard history & snippets, TTL cache, file-organizer overlay, and a knowledge graph — plus a live desktop activity lens. 40 tools, SQLite everyw
0xsline/OpenChatCutLocal-first conversational AI video editor with a professional multi-track timeline, Agent Skills, MCP integration, and Remotion-powered rendering.
herry2059/project-os-for-codexOpen-source control plane for Codex projects: Git-backed context, visible agent progress, scoped MCP access, resumable work, and safe handoffs.
calesthio/ResonantFree, local AI music studio for Windows—generate songs, play instruments, arrange, mix, export WAV, and connect Codex or Claude through MCP.
NLACE-COM/mcp-supermercados-clServidor MCP para buscar productos, comparar precios y armar la mejor lista de compra en supermercados chilenos (Jumbo, Santa Isabel, Unimarc, Tottus, Lider) desde Claude o ChatGPT. 100% local.
janwilmake/agent-codemodeLet the scripts your coding agent writes call your connected MCP servers (no extra auth)
The top new MCP servers of the week, every Monday. No spam, unsubscribe anytime.