SkillAgentSearch skills...

open-fusion-mcp

MCP server and agent skills for motion design in DaVinci Resolve Fusion: 179 checked operations, a one-call scene builder and a render-and-look loop.

Install / Use

claude mcp add legionsound -- npx -y github:legionsound/open-fusion-mcp

If the server publishes to npm under a different name, use that package instead — check the repo README.

About this skill
🔌

MCP Server

Model Context Protocol server

Quality Score

81/100

Supported Platforms

Claude Code
Claude Desktop

Our assessment of open-fusion-mcp

open-fusion-mcp scores 81/100 on our quality scale, 3499th of 4,589 Development & Engineering skills we index.

Its MCP Server is 10 KB long, well organised into 12 sections with 3 code examples: a thorough specification that gives an agent plenty to work with.

It has 10 GitHub stars, so there is little community track record yet; judge it on its content.

Substance
29/30
Structure
18/20
Description
15/15
Adoption
4/20
Freshness
15/15

Maintenance, license and trust

  • The repository was last updated 21 days ago, so open-fusion-mcp is actively maintained.
  • It is released under the MIT license, a permissive license that allows use, modification and commercial use with attribution.
  • Its trust signals score 97/100, with no cautions. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.

open-fusion-mcp compared with similar skills

All 4 of these similar skills score higher than open-fusion-mcp; compare them before choosing.

SkillScoreStarsUpdatedFormat
open-fusion-mcp (this skill)by legionsound811021d agoMCP Server
taste-skillby Leonxlnx100101.4k2d agoCLAUDE.md
Agent-Reachby Panniantong100101.2ktodayCLAUDE.md
headroomby headroomlabs-ai10077.6k1d agoCLAUDE.md
ai-job-searchby ai-job-search-dev10063.9k2d agoCLAUDE.md

Frequently asked questions

How do I install open-fusion-mcp?
Run claude mcp add legionsound -- npx -y github:legionsound/open-fusion-mcp. The install tabs above show the steps for each supported agent.
Which AI agents does open-fusion-mcp work with?
It is written for Claude Code and Claude Desktop, as a MCP Server file. Other agents that read the same format can often use it too.
Is open-fusion-mcp safe to use?
It is MIT-licensed and scores 97/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
Is open-fusion-mcp still maintained?
The repository was last updated 21 days ago, so open-fusion-mcp is actively maintained.
<div align="center">

open-fusion-mcp

Agent-built motion design, native in DaVinci Resolve Fusion.

An MCP server and a set of agent skills that let an AI agent design, build, render and check real Fusion node graphs: editable titles, kinetic type, UI animation and whole multi-scene films.

License: MIT CI Python 3.12 DaVinci Resolve Studio 21.1 MCP macOS

https://github.com/user-attachments/assets/c0c83852-3371-4a8f-9411-3cb59df8130c

Every frame of this film was built by an agent in Fusion, as one editable comp.

Quick start · How it works · What's inside · Results · Docs

</div>

Why

Fusion is one of the most capable compositing and motion tools there is, and it ships inside DaVinci Resolve. Agents have had a hard time driving it: one After Effects layer becomes three to ten Fusion nodes, units differ from input to input, and silent failures are easy. open-fusion-mcp closes that gap with a fast, checked path into Fusion and the craft knowledge an agent needs to use it well. The connector is a fast path, not a cage: agents combine it with the official DaVinci Resolve MCP and computer use as one toolbox.

Made with it

The film above is the project's own explainer. It is 59 seconds and 1,779 frames across seven scenes, built as one Fusion comp of about 9,000 native nodes, with 3D cameras, 12-pass depth of field and motion blur. Claude Code agents built every frame through this toolkit from a written brief, and a human editor directed it through notes on the edit timeline over twelve review rounds. There is no stock footage and no template: the titles, shapes, cards, cameras and node-graph visuals are Fusion tools the agent wrote.

Highlights

  • 181 checked operations in 36 categories behind 15 MCP tools (fu_*): comps, tools, inputs, keyframes, expressions, masks, text, 3D, templates, timelines, renders and Deliver. Every write is read back.
  • One-call scene builder. Describe layers, type styles, layout, enter/exit moves and a camera; get an editable native graph (scene.build), then change it by description (scene.update). Multi-scene films live in one culled comp on a "film ladder".
  • Render-and-look loop. Frame renders with inline previews, contact sheets, image comparison and a motion audit, so the agent checks pixels instead of trusting return values.
  • Build-loop speed. Disk caches for finished branches (-42 % per render), a one-call graph tidier (560 tools in 1.3 s) and a draft/final switch that flips a whole film in one control.
  • Guard rails. Project allowlist, read-only mode, category allowlist, dry runs and checked connections, with clear refusals instead of half-applied edits.
  • Recovery after a timeout. The worker journals each operation step by step, so the reply to a timeout or a dead worker carries a receipt: what finished, what was running and what never started. batch.recover re-reads the comp, batch.rollback undoes the call and verifies the result, resume finishes the batch without repeating a step, and atomic: true makes a batch all or nothing. Fault-injection tests against a fake Resolve cover these paths, and a live check in a scratch project confirmed them in Resolve Studio 21.1.
  • Craft skills. A motion-design procedure and router, a distilled Fusion 21.1 reference and a Figma-to-Fusion transfer, written for agents and served by the MCP server itself.

How it works

flowchart LR
    A["AI agent<br/>(Claude Code or any MCP client)"] -->|"fu_* tools"| S["use-fusion MCP server"]
    S -->|"skills on request"| K["skills/<br/>craft, reference, transfer"]
    S --> W["worker process<br/>(one Resolve caller)"]
    W -->|"Fusion scripting API"| R["DaVinci Resolve Studio 21.1<br/>Fusion page"]
    W -.->|"one line per step"| J["call journal<br/>(receipts, recovery)"]
    A -.->|"full Resolve API"| O["official DaVinci Resolve MCP"]
    A -.->|"dialogs, UI-only controls"| C["computer use"]
    O -.-> R
    C -.-> R

The agent loads the use-fusion skill first, looks up operations with fu_catalog (one operation, a keyword search or a category), acts with fu_do (a batch is one undo step; atomic: true makes it all or nothing), and verifies with fu_render_frame. The most-used operations also have their own tools, fu_scene_build, fu_scene_plan, fu_batch and fu_contact_sheet, with schemas generated from the operation definitions, and every bad-argument reply lists the operation's parameters, so a retry takes one call. If Resolve stops answering, the reply carries a receipt from the call's journal, and the batch can be rolled back or resumed. Details: architecture.

Quick start

Requirements: macOS (tested on Apple Silicon), DaVinci Resolve Studio 21.1 with Preferences > System > General > External scripting using = Local, Python 3.12 and an MCP client. The free edition of Resolve does not allow external scripting.

git clone https://github.com/legionsound/open-fusion-mcp.git
cd open-fusion-mcp/connector
python3.12 -m venv .venv && .venv/bin/pip install -r requirements.txt
bin/fusion-connector doctor      # checks Python, the scripting port, Resolve, Studio, skills and data tables
bin/fusion-connector config      # prints the registration command for your MCP client

Register the server (Claude Code shown; restart the session afterwards):

claude mcp add use-fusion --scope user \
  -e FUSION_MCP_PROJECT_ALLOWLIST=Testbed \
  -- /absolute/path/to/open-fusion-mcp/connector/bin/use-fusion-mcp

Then generate the Fusion data tables from your own Resolve install (how), open a scratch project named Testbed, and ask your agent to call fu_version_info and load the use-fusion skill. Full guide: docs/install.md.

[!TIP] Keep FUSION_MCP_PROJECT_ALLOWLIST set while you learn the tools: every mutating operation refuses projects outside the list, and reads still work everywhere.

What's inside

| Path | What it is | |---|---| | connector/ | the use-fusion MCP server (Python 3.12), its CLI, offline and live tests, and PARITY.md (After Effects connector operations mapped to Fusion) | | skills/ | agent skills, served by the server and usable directly by your agent | | agents/ | example Claude Code agent definitions: a builder, a connector engineer and a performance researcher | | docs/ | install, architecture, toolbox, scene builder, caching, graph layout, results, fonts, maintaining | | scripts/ | the maintainer's sync and audit tools |

| Skill | Use it for | |---|---| | use-fusion | the connector's entry point: working loop, operations, safety, known limits | | fusion-motion-design | the creative procedure and router: titles, kinetic type, UI, glass, depth, transitions, characters, collage, delivery | | fusion-reference | Fusion 21.1 facts for agents: nodes, inputs and IDs, .setting format, measured "realities" of the API | | fusion-figma-transfer | Figma frames into editable native Fusion nodes (needs a Figma connection) |

The film mentions six skills. Two VFX skills, fusion-cleanup and fusion-matte-painting, are being rewritten and will join in a later release.

Measured results

Measured on one Mac (Apple Silicon, 32 GB) with Resolve Studio 21.1. The benchmark is a 25-second 3D SaaS ad, first built in After Effects and then rebuilt cold from the same written spec.

| Build | Agent time | Tool calls | |---|---|---| | After Effects (reference) | about 1 h 55 min | about 200 | | Fusion, first build without the scene builder | about 5.6 h | about 830 | | Fusion with the scene builder, one comp | 1 h 19 min | about 156 |

With the scene builder an agent builds a 2,911-tool Fusion film in about the time the After Effects build took, with a fifth of the first Fusion build's tool calls. More numbers and caveats: docs/results.md.

Documentation

| Guide | | |---|---| | Install | requirements, registration, permissions, policy switches, troubleshooting | | Architecture | server, worker, receipts and recovery, operations, skills store, tests | | The toolbox | when to use this connector, the official Resolve MCP or computer use | | Scene builder | the scene description format, films, updates | | Caching | disk caches for the build loop, staleness and refresh | | Graph layout | the tidier and the house graph style | | Results | benchmarks, render efficiency, tests | | Fonts | which fonts the examples use | | Maintaining | how this repository is synced and audited |

Status

Version 0.2. Tested on macOS (Apple Silicon) with DaVinci Resolve Studio 21.1.0.14, Python 3.12 and Claude Code. Windows and Linux are untested; the server is plain Python, but the helper that dismisses Resolve's render dialogs uses AppleScript. See CHANGELOG.md for what is new.

Contributing

Issues and pull requests are welcome. Start with CONTRIBUTING.md; run the offline tests before you open a pull request. Please report security issues privately as described in SECURITY.md.

License and credits

MIT, Copyright (c) 2026 Legion Media LLC. See LICENSE.

This project is the Fusion counterpart of Higgsfield's After Effects connector and skills, and it learned from several open-source DaVinci Resolve MCP servers; CREDITS.md lists them with their licenses. DaVinci Resolve and Fusion are trademarks of Blackmagic Design Pty. Ltd. This project is not affiliated with or endorsed by Blackmagic Design or Higgsfield.

Related Skills

View on GitHub
GitHub Stars10
CategoryDevelopment
Updated21d ago
Forks1

Languages

Python

Trust signals

97/100

From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.

1 info