resolve-mcp
Provides an AI assistant with 132 tools to control DaVinci Resolve Studio on macOS, enabling project management, media handling, timeline editing, color grading, rendering, and Fusion compositing through natural language commands.
README
resolve-mcp
An MCP server that gives an AI assistant control of DaVinci Resolve Studio on macOS.
132 tools across projects, media, timelines, colour, rendering and Fusion — plus
resolve_exec for anything the curated tools miss, and resolve_ui for the handful of
operations Blackmagic never exposed to scripting at all.
"Import this folder, build a timeline in filename order, add a marker at every
scene cut, then queue ProRes and H.264 renders and tell me when they're done."
Install
Requires macOS, DaVinci Resolve Studio (external scripting is a Studio feature), Python 3.13+ and uv.
git clone https://github.com/irstone-source/resolve-mcp.git
cd resolve-mcp
uv sync
uv run python scripts/smoke.py # verify the connection before wiring anything up
Then register with Claude Code:
claude mcp add resolve -- uv run --directory /absolute/path/to/resolve-mcp resolve-mcp
One manual step, once. In Resolve: Preferences → System → General → External scripting using:
Local. No API can set it, and without it nothing works. Full detail in docs/SETUP.md.
Documentation
| Setup | Requirements, permissions, headless mode |
| Tool reference | All 132 tools and 6 resources, generated from the server |
| Architecture | How it works and why it's built this way |
| Troubleshooting | Failure modes and Resolve's undocumented behaviour |
| Development | Adding tools, testing, conventions |
How to drive it
Reach for capabilities in this order:
- A curated tool —
project_*,media_*,timeline_*,track_*,marker_*,clip_*,lut_*,render_*,fusion_*. Typed, validated, with actionable errors. resolve_exec— arbitrary Python against the live connection, for the long tail of the API. The installed scripting reference is served as MCP resources, so calls are written against real signatures rather than recalled ones.resolve_ui— AppleScript menu clicks and keystrokes. Only where no API exists.
Handles
Resolve returns live objects that can't be sent as JSON, so clips, timeline items, folders and stills pass between calls as handles:
clip:1a2b3c a media pool clip
item:0091 a clip on a timeline
folder:root a bin
timeline:4f2a a timeline
still:2b70 a gallery still
Objects returned by resolve_exec get handles too, so escape-hatch code composes with the
curated tools.
Every tool returns the same envelope
{ "ok": true, "data": { ... } }
{ "ok": false, "error": "No timeline is currently open",
"hint": "Create a timeline with timeline_create, or select one with timeline_set_current." }
The Resolve API signals failure by returning bare False or None with no explanation.
This server converts each known cause into a named condition with a remedy, so a failed
call tells you what to do next instead of merely that something went wrong.
What it's good at
Repetitive and generative work: batch operations across many clips or timelines, conforming, marker-driven assembly, and querying an edit as data. It reads a timeline as structured information — every clip with its in/out, duration, track position and metadata.
What it can't do is exercise visual judgement. It sees metadata, not pictures, unless you ask it to render a still and look at one.
Notable behaviour it works around
scriptapp()blocks forever when Resolve is running but refuses a connection — inside C, without releasing the GIL, so no signal or thread timeout recovers it. Every connection is probed in a killable subprocess first.AddRenderJobreturns'', notFalse, when it declines — so an empty job id would otherwise be handed back as success.- Resolve's dialogs are invisible to macOS Accessibility (they're Qt-drawn), and a modal blocks scripting while open. Tools validate inputs client-side rather than provoking one.
- Unknown render setting keys are silently ignored, making a typo look like success. Keys are validated against the installed reference.
More in docs/TROUBLESHOOTING.md.
Development
uv run pytest # 195 unit tests, no Resolve needed
uv run pytest -m live # integration suite against real Resolve
uv run ruff check . && uv run ruff format .
Unit tests run against an in-memory double shaped like the real API, so the suite needs no
Resolve installation and no application state to reset. The live suite works inside a
throwaway MCP_TEST_<pid> project and restores whatever project was open.
Status
Verified against DaVinci Resolve Studio 21.0.2.4 on macOS. 195 unit tests and 13 live integration tests passing.
Licence
MIT — see LICENSE.
Not affiliated with or endorsed by Blackmagic Design. DaVinci Resolve is a trademark of Blackmagic Design Pty Ltd.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。