Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@RV MCP ServerLoad shot_v02 and compare it with v01 using a wipe"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
RV MCP Server
MCP (Model Context Protocol) server that bridges AI assistants like Claude to Autodesk/Tweak RV, the industry-standard media review application. Control playback, compare shots, adjust color grading, and manage review sessions — all through natural language.
No plugin required inside RV. Uses RV's built-in network listener with Mu scripting via remote-eval.
Requirements
RV 2022.3.1+ with network mode enabled (
-networkflag)Python 3.10+
uv package manager
Quick Start
1. Start RV with networking
rv -networkRV will listen on port 45124 (default). On Windows:
"C:\Program Files\ShotGrid\RV-2022.3.1\bin\rv.exe" -network -networkPort 451242. Install and register
Claude Code (CLI):
claude mcp add --scope user rv-mcp -- uv run --directory /path/to/RV_MCP rv-mcpClaude Desktop (~/.claude.json):
{
"mcpServers": {
"rv-mcp": {
"command": "uv",
"args": ["run", "--directory", "/path/to/RV_MCP", "rv-mcp"]
}
}
}3. Use it
Ask Claude to load media, control playback, compare shots, or adjust colors. The server translates natural language into RV commands automatically.
Architecture
Claude (stdio/MCP) --> FastMCP Server --> RV Network Protocol (TCP:45124) --> RVThe server maintains a persistent TCP connection to RV using a custom protocol based on RV's RvCommunicator. Key design decisions:
Persistent connection with automatic reconnection on socket loss
Thread-safe via
threading.Lockfor concurrent tool callsClean shutdown via
atexithandler that sendsDISCONNECT(without this, RV rejects future connections)Mu string handling — return values are automatically unquoted and unescaped
Protocol Flow
1. Connect TCP to 127.0.0.1:45124
2. Send: NEWGREETING <len> rv-mcp rvController
3. Send: PINGPONGCONTROL 1 0 (disable heartbeat)
4. Recv: NEWGREETING <len> <rv-name> (consume RV's greeting)
5. For each command:
Send: MESSAGE <len> RETURNEVENT remote-eval * { require commands; <mu_code> }
Recv: MESSAGE <len> RETURN <value>
6. On shutdown:
Send: MESSAGE <len> DISCONNECTTools (41 total)
Execute (1)
Tool | Description |
| Run arbitrary Mu code — escape hatch for anything not covered by dedicated tools |
Playback (17)
Tool | Description |
| Load a media file (image sequence, movie, or single image) |
| Load multiple media files at once |
| Start playback |
| Stop playback |
| Toggle play/stop, returns new state |
| Get current frame number |
| Jump to a specific frame |
| Step forward by N frames (default 1) |
| Step backward by N frames (default 1) |
| Set the in-point (start of playback range) |
| Set the out-point (end of playback range) |
| Get current in/out points as JSON |
| Set playback frames per second |
| Get current playback FPS |
| Enable/disable realtime mode (skip frames to maintain FPS) |
| Set loop mode: |
| Set playback direction and speed (1=forward, -1=reverse, 2=2x, etc.) |
| Get full playback state as JSON (frame, range, in/out, playing, fps) |
Sources (7)
Tool | Description |
| List all loaded source nodes as JSON array |
| Get detailed media info (resolution, frame range, fps, bit depth, channels) |
| Get source nodes visible at a specific frame |
| Create a new empty session |
| Clear all sources from the current session |
| Save session to an |
| Get session state as JSON (view node, frame range, source count) |
Compare (4)
Tool | Description |
| Switch view: |
| Set stack composite mode: |
| Toggle A/B wipe comparison (auto-switches to stack view) |
| Get current view state as JSON |
Color (12)
Tool | Description |
| Load a LUT file ( |
| Deactivate LUT on a target |
| Set CDL values (slope, offset, power, saturation) — partial updates supported |
| Deactivate CDL color correction |
| Set exposure (per-channel or uniform) |
| Set gamma correction |
| Set saturation |
| Get current color correction state as JSON |
| Set display gamma (e.g., 2.2 for sRGB-like) |
| Enable/disable sRGB display transform |
| Set viewport background: |
Usage Examples
Load and review footage
"Load the EXR sequence at /shots/sh010/comp/sh010_comp.1-100#.exr"
"Play it back at 24fps"
"Go to frame 50"
"Set in point at 20 and out point at 80"Compare two versions
"Load both /shots/sh010/comp_v1.mov and /shots/sh010/comp_v2.mov"
"Switch to stack view"
"Set composite to difference mode"
"Toggle the wipe to compare side by side"Color correction
"Apply CDL with slope [1.1, 0.95, 1.0] and saturation 1.2"
"Load the ACES LUT from /luts/sRGB.cube"
"Set exposure to 0.5"
"Show me the current color settings"Advanced (raw Mu)
"Execute this Mu code: { require commands; let s = sources(); string(s.size()); }"Project Structure
RV_MCP/
├── pyproject.toml # Package config, entry point, dependencies
├── README.md
├── .gitignore
└── src/
├── __init__.py
├── server.py # FastMCP server + RvClient instantiation
├── rv_client.py # Persistent TCP client (RV network protocol)
└── tools/
├── __init__.py
├── execute.py # execute_mu — raw Mu escape hatch
├── playback.py # 17 playback/transport tools
├── sources.py # 7 source & session tools
├── compare.py # 4 view/compare tools
└── color.py # 12 color/LUT/CDL toolsTroubleshooting
"Could not connect to RV"
Ensure RV is running with the
-networkflagCheck that port 45124 is not blocked by a firewall
Use
-networkPort 45124to explicitly set the port
RV rejects connections after a crash
If the server exits without sending DISCONNECT, RV may reject new connections. Restart RV to clear the state. The server includes an atexit handler to prevent this under normal operation.
Mu code errors
Always wrap code blocks in
{ require commands; ... }Mu evaluates both branches of
if/then/else— avoid property access on nodes that may not existFile paths must use forward slashes;
escape_mu_string()handles this automatically
Timeout errors
The default timeout is 30 seconds. If Mu code takes longer (e.g., loading large sequences), it may time out. Use execute_mu for long operations and consider breaking them into smaller steps.
Development
# Install dependencies
uv sync
# Run the server directly
uv run rv-mcp
# Run with debug logging
uv run python -m src.serverLicense
MIT
Resources
Looking for Admin?
Admins can modify the Dockerfile, update the server description, and track usage metrics. If you are the server author, to access the admin panel.