davinci-resolve-lite-mcp

davinci-resolve-lite-mcp

Enables AI clients such as Claude Code to control DaVinci Resolve, including the free Lite edition, through a local HTTP server that runs inside Resolve's Scripts menu. It exposes 163 tools for editing, color, rendering, media pool, and Fusion title styling.

Category
Visit Server

README

davinci-resolve-lite-mcp

test PyPI MCP Registry License: MIT Python 3.9+ Platform: macOS DaVinci Resolve: Lite  Studio 163 tools Zero dependencies

https://github.com/user-attachments/assets/8429932f-643b-4131-bdf6-dad0d3399137

Claude builds an opening title in DaVinci Resolve Lite — a gold "GameHelper" Text+ node with glow and a zoom-in keyframe reveal — from a plain-language request, via insert_fusion_title + style_fusion_title.

An MCP server that lets an AI client such as Claude Code control DaVinci Resolve — including the free (Lite) edition, which the existing davinci-resolve-mcp project cannot drive.

The free edition blocks external scripting, but it still runs Python scripts launched from its own Workspace > Scripts menu. This project rides that path: the MCP server runs inside Resolve as a menu script, and exposes Resolve's Python API over a small local HTTP endpoint that Claude connects to.

Claude Code ──HTTP JSON-RPC (MCP)──▶  127.0.0.1:8765/mcp
                                          │   server runs INSIDE Resolve
                                          │   (Workspace > Scripts > Utility)
                                          ▼
                              command queue → main script thread
                                          ▼
                              global `resolve` object → Resolve API

Tools

Ask Claude in plain language; it drives Resolve through the tools — see the demo above. See the tools reference for the full 163-tool surface — editing, color, render, media pool, and Fusion title styling.

Why this works on the free edition

  • Free Resolve permits scripts run from its Scripts menu (only external network scripting is restricted).
  • A menu script gets the resolve object for free and may run a long-lived loop — long enough to host a server.
  • The sandboxed Lite app ships the com.apple.security.network.server entitlement, so it can open a localhost listening socket.
  • Zero dependencies — pure Python standard library. Nothing to pip install into Resolve's interpreter.

Requirements

  • macOS with DaVinci Resolve (Lite/free or Studio).
  • Claude Code (or any MCP client that speaks the Streamable HTTP transport).

Install

git clone https://github.com/2sem/davinci-resolve-lite-mcp.git
cd davinci-resolve-lite-mcp
./install.sh

Or via pip, if you'd rather not clone the repo:

pip install davinci-resolve-lite-mcp
davinci-mcp-install

davinci-mcp-install does exactly what install.sh does (same Lite/Studio detection, same copy-vs-symlink logic) — it just reads the files from your pip-installed package instead of a git checkout. davinci-mcp-uninstall reverses it. Either way, the sandbox note below still applies, and you still start the server from Resolve's own menu — see Run.

macOS user-install note. If you see Defaulting to user installation because normal site-packages is not writeable, pip installed the console scripts under $(python3 -m site --user-base)/bin, which usually isn't on PATH. Either run the full path — "$(python3 -m site --user-base)/bin/davinci-mcp-install" — or add that bin directory to PATH. And use &&, not &, between the two commands: a bare & backgrounds pip install and races davinci-mcp-install before the package exists.

Also listed on the MCP Registry as io.github.2sem/davinci-resolve-lite-mcp (verify) — for discoverability only. The listing is metadata-only (no auto-install packages/remotes entry): this server can't be spawned by an MCP client the way a typical registry server can, since it must run inside Resolve's own embedded Python interpreter, started by hand from the Scripts menu. Install it one of the two ways above.

install.sh deploys:

  • the two launcher scripts into Fusion/Scripts/Utility (a folder Resolve scans for the Scripts menu; Utility shows on every page), and
  • the resolve_mcp package into Fusion/Scripts/MCP — a folder Resolve does not scan, so the helper modules stay out of the menu.

The Lite container path is detected automatically.

Sandbox note (important). DaVinci Resolve Lite is sandboxed and can only read its own container, ~/Movies, and files you pick interactively. A symlink that points outside those locations (e.g. into a clone under ~/Projects) cannot be followed by the sandboxed app, so the menu script would silently never run. For that reason install.sh copies the files into the container on Lite (and symlinks only on the non-sandboxed Studio build). Re-run ./install.sh after pulling updates.

Resolve enumerates only the category folders (Utility / Comp / Tool / Edit / Color / Deliver) for its Scripts menu — that is why the launchers go in Utility and the package hides in MCP.

The same sandboxing applies to file paths you ask the tools to use: exports/imports should target ~/Movies (or other granted locations), otherwise Resolve cannot write/read them.

Run

  1. In DaVinci Resolve: Workspace > Scripts > Utility > davinci_mcp_server.

    Workspace > Scripts menu showing davinci_mcp_server and stop_davinci_mcp_server

  2. Open Workspace > Console — it prints the endpoint and port:

    MCP endpoint:  http://127.0.0.1:8765/mcp
    Add to Claude Code:
      claude mcp add --transport http davinci http://127.0.0.1:8765/mcp
    

    The startup guide prints to the Resolve Console (Workspace > Console). Because the server runs continuously, its Console output can buffer until it stops, so both scripts also mirror every line to a logfile:

    ~/Movies/davinci-resolve-lite-mcp.log
    

    Watch it live with ./logs.sh. (Override the directory with DAVINCI_MCP_LOG_DIR.) ~/Movies is used because the sandboxed Lite app is allowed to write there.

    Once running, every tool call is logged to the Console as a single line ([davinci-mcp] <name> <args> -> ok|error (Nms)):

    Resolve Console showing per-command davinci-mcp log lines

    Update check. Like brew/CocoaPods, each launch checks PyPI in the background for a newer release and prints a one-line nudge to the Console if one exists — never blocks startup, and any failure (offline, PyPI down) stays silent in the logfile only. Disable with DAVINCI_MCP_SKIP_UPDATE_CHECK=1.

  3. Register it with Claude Code (one-time):

    claude mcp add --transport http davinci http://127.0.0.1:8765/mcp
    

    Then verify / reconnect with the /mcp command inside Claude Code — it lists connected servers and reconnects them. If Claude was already running when you launched the script, type /mcp (or restart the session) so it picks up the davinci server.

  4. Ask Claude to control Resolve.

Configure the port (stable, recommended)

By default the server listens on 8765 and auto-increments to 8766, 8767, … if that port is busy (another local tool may already hold 8765). Because the winner of that race can change between launches, the URL you registered with Claude can drift, surfacing as:

Failed to reconnect to davinci: HTTP 404 at http://127.0.0.1:8765/mcp

To lock the port for good, drop a small JSON config file. When a port is set this way it is pinned — the server binds exactly that port and never auto-increments, so you register Claude once and the URL never moves.

Create ~/Movies/davinci-resolve-lite-mcp.config.json:

{ "host": "127.0.0.1", "port": 8770 }

Why ~/Movies and not ~/.config? The Lite app is sandboxed and can only read its own container, ~/Movies, and files you pick interactively — ~/.config is outside the sandbox, so Lite cannot read it (this is the same reason the logfile lives in ~/Movies). The server also checks ~/.config/davinci-resolve-lite-mcp/config.json for the non-sandboxed Studio build, where that path is conventional.

Then restart the server (Scripts > Utility > stop_davinci_mcp_server, then davinci_mcp_server) and register Claude once at the fixed port:

claude mcp add --transport http davinci http://127.0.0.1:8770/mcp

The Console banner confirms the source — look for Port : pinned (from …) — will not auto-increment.

Resolution order (highest priority first): the DAVINCI_MCP_PORT / DAVINCI_MCP_HOST environment variables, then the config file, then the built-in defaults. The env vars also pin the port, but a Dock-launched Resolve won't see a shell export; the config file is the simplest persistent option. DAVINCI_MCP_CONFIG=/path/to.json forces a specific config file exclusively — if that path is missing or malformed the server falls back to the built-in defaults rather than reading ~/Movies / XDG.

If the port already drifted and Claude points at the wrong one, re-point it:

claude mcp remove davinci
claude mcp add --transport http davinci http://127.0.0.1:<actual-port>/mcp

Stopping

Any of these stops the server:

  • From the menu: Workspace > Scripts > Utility > stop_davinci_mcp_server
  • From a terminal: ./stop.sh
  • Quit DaVinci Resolve

The menu stop script and stop.sh both POST to the server's /shutdown endpoint, scanning the same port range the server uses on startup.

The port auto-increments from 8765 only when it is not pinned. To lock it so the URL never moves between launches, see Configure the port.

Tools

163 tools, spanning the full pipeline:

  • Status & navigation — page switching, project/timeline settings
  • Projects & timelines — load/create/duplicate, markers, scene cuts, lifecycle
  • Tracks — add/delete, enable/lock/rename
  • Editing — place/append/delete clips, titles & generators, transform/crop/zoom
  • Media pool & storage — import/delete, properties & metadata, tagging, disk browse
  • Color — node graph LUT/enable, reset grades, stills
  • Render & export — render queue, formats/codec, frame/timeline/project export & import

See docs/TOOLS.md for the complete per-tool reference.

Every tool call is logged to the Resolve Console and the logfile as a single line: [davinci-mcp] <name> <args> -> ok|error|EXCEPTION (Nms).

Project layout

src/davinci_mcp_server.py        thin launcher (deployed to Scripts/Utility)
src/stop_davinci_mcp_server.py   stop launcher
src/resolve_mcp/                 the server package (deployed to Scripts/MCP, hidden)
    config · logio · connection · bridge · tools · server
tests/test_server.py             offline tests (fake Resolve, no app needed)
install.sh · uninstall.sh · stop.sh · logs.sh
docs/TOOLS.md                    full per-tool reference
fallbacks/                       documented gotchas + fixes

Testing

  • Offline (no Resolve, no server) — import + dispatcher + tool-count smoke:
    python3 tests/test_server.py
    
  • Live integration — one test per tool against a running server (Resolve open with a project + a media clip, and davinci_mcp_server launched):
    python3 tests/live_test.py                 # all features
    python3 tests/live_test.py set_timecode    # run the test(s) for given feature(s)
    
    Each test name equals the tool name, so when you change a tool you can run just its test: python3 tests/live_test.py <tool>. Tests are reversible (scratch timeline + temp files, cleaned up). File-dependent and session-destructive tools are checked via their error path; Studio-only / heavy tools (e.g. detect_scene_cuts, render_current_timeline, quick_export) are skipped with a reason. A few marker / still tests depend on a clean Resolve session state — re-run them after a fresh launch if they flake.

Contributing

See CONTRIBUTING.md for how to add a tool, run the suites, the stdlib-only / Lite-first constraints, and the release flow.

Scope

This server targets the free (Lite) edition and intentionally covers only API that runs there. Studio-only / paid features are deliberately omitted (they no-op or error on Lite), namely: audio transcription, subtitles-from-audio, Magic Mask, Stabilize, Smart Reframe, Dolby Vision analysis, Voice Isolation, and cloud projects / database management. The remaining unwrapped methods are trivial accessors (GetUniqueId, cache modes, Fusion-comp internals, takes, stereo/3D, layout & burn-in presets, mattes) — not functional gaps.

Known limitations

  • Clips can be addressed by name (within the current media-pool folder) or by id (id/ids, resolvable across any bin) — pass id/ids when names are ambiguous or the clip lives in another folder.
  • Tool arguments are validated against each tool's JSON Schema (required fields, basic types, and enums); a malformed call returns a clear error naming the offending argument. Deep/nested schema constraints are not exhaustively checked.

Security note

The server binds to 127.0.0.1 only, so it is reachable from your machine only. It exposes control of DaVinci Resolve to any local process that can reach the port — only run it on a machine you trust.

The one outbound call the server makes on its own is the startup update check (a GET to PyPI's public JSON API for the current version number — no other data sent). Disable it with DAVINCI_MCP_SKIP_UPDATE_CHECK=1 if you'd rather it made none.

License

MIT — see LICENSE.

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured