renderdoc-mcp
An MCP server and CLI for RenderDoc: an agent opens GPU frame captures and inspects draw calls, shaders and pixels
- 373 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The server runs an .exe with the RenderDoc library and reads capture files on your computer
- Export writes textures, buffers and meshes to disk, and live capture hooks a running application
Install
- Works in
Manual install
{
"mcpServers": {
"renderdoc-mcp": {
"command": "renderdoc-mcp.exe",
"args": []
}
}
}Give the full path to renderdoc-mcp.exe if the bin folder is not in PATH.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The server runs an .exe with the RenderDoc library and reads capture files on your computer
- Export writes textures, buffers and meshes to disk, and live capture hooks a running application
This is third-party code. Review the repository files before installing.
What it does
renderdoc-mcp wraps the RenderDoc replay API in 59 structured tools. The agent opens .rdc files, walks events and draw calls, and reads pipeline state together with shaders. No manual work in the RenderDoc interface is needed, and a frame can also be captured live. Export gives render targets, textures, buffers and meshes as evidence for a report. Comparing two captures and asserting pixels let you wire a graphics check into a build. The server speaks stdio, and the renderdoc-cli command-line tool sits next to it for shell and CI. The release archive also carries a renderdoc-mcp skill for Codex.
Who it is for. For graphics programmers and testers of games and 3D applications who want an agent to analyze frames in RenderDoc.
Good fit when
- A frame renders wrongly and you need the draw call that spoiled a pixel
- You need to compare two captures before and after a fix and show where state changed
- You need the shader, constant buffers and bindings of a specific call
- You need automatic pixel and state assertions in CI for a graphics test
- You need to export a render target, texture or mesh as evidence for a bug report
Not a fit when
- You need to drive a game engine: Unity, Godot and Unreal have their own MCP servers
- You need GPU performance profiling: the server reads captures and state and does not replace a profiler
- You are not on Windows and not ready to build from source: the prebuilt releases target Windows
- You have no .rdc captures and no application you can capture
Example request
Open capture.rdc, find the draw call that painted pixel (512, 300) red and show the pipeline state and the shaderMCP
- Transport
- stdio
- Authentication
- not required
Limitations
The prebuilt archive on GitHub Releases holds Windows .exe files: renderdoc-mcp.exe, renderdoc-cli.exe, the renderdoc.dll library with renderdoc.json and the install-codex.ps1 installer for Codex Desktop. For another platform you build the server from source with CMake, pointing the RENDERDOC_DIR variable at the RenderDoc source root, plus RENDERDOC_BUILD_DIR if the RenderDoc build is in a non-standard place.
RenderDoc itself is distributed under its own licence, separate from this project's MIT.
How to disable. Remove the renderdoc-mcp entry from the client config: the [mcp_servers.renderdoc-mcp] section in ~/.codex/config.toml for Codex, or the renderdoc-mcp object under mcpServers for Claude Code. Then delete the unpacked folder with the .exe files.
FAQ
How do I connect renderdoc-mcp to Claude Code?
Download the archive from GitHub Releases, unpack it and add a renderdoc-mcp entry to mcpServers with command renderdoc-mcp.exe and an empty args list. The agent then sees the RenderDoc tools.
What is in the release archive?
A bin folder with renderdoc-mcp.exe (the stdio server), renderdoc-cli.exe (a CLI for shell and CI) and the renderdoc.dll and renderdoc.json libraries, a skills/renderdoc-mcp folder with a Codex skill and the install-codex.ps1 installer.
What jobs do the tools cover?
Open an .rdc or capture a frame live, jump to any event, inspect pipeline state, bindings, shaders and constant buffers, analyze passes and resource usage, get pixel history and debug a pixel, vertex or thread, and export render targets, textures, buffers and meshes.
Can I use it in CI?
Yes, there is a diff and assertions block. It compares two captures and asserts pixels, state and images, and renderdoc-cli runs from a shell. That lets you catch graphics regressions without manual review.
How do I build the server from source?
Run cmake -B build -DRENDERDOC_DIR=<path to RenderDoc source>, then cmake --build build --config Release. If the RenderDoc build is in a non-standard place, set RENDERDOC_BUILD_DIR. You need to obtain the RenderDoc source separately.
Discussion
No questions or reviews yet.
Related
MCP serversAn MCP server that builds a code graph once and serves the answer to the agent instead of re-reading files
An MCP server for draw.io: the agent builds and edits diagrams, works with layers, pages and cloud provider shapes
An MCP server that lets an agent read files, search code and run commands in a workspace
An MCP server that indexes a repository into a symbol graph and hands agents cards, slices and code windows under a context budget
Docs MCP Server
Grounded Docs MCP Server
A local MCP server that indexes docs from websites, GitHub, npm, PyPI and files and gives the agent version-aware search
MCP Server for WinDbg
MCP Server for WinDbg Crash Analysis
An MCP server that hands the agent the Windows debuggers: crash dump analysis, remote and kernel debugging in natural language
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| renderdoc-mcp (this entry) | no | not required | 373 | |
| MCP for Unity | no | not required | 15,537 | drives the Unity editor: scenes, assets, scripts; it does not analyze GPU frames or shaders |
| Ghidra MCP Server | no | not required | 5,976 | analyzes binaries through Ghidra; renderdoc-mcp reads GPU frame captures |
README in short
Capabilities sit in a table across seven areas, from frame navigation to CI assertions. Codex Desktop and Claude Code configuration and the source build get their own sections, and a diagram shows the architecture.