RenderDoc MCP Server
Click on "Deploy 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., "@RenderDoc MCP Server List the draw calls in the currently loaded capture."
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.
RenderDoc MCP Server
An MCP server that runs as a RenderDoc UI extension. It allows AI assistants to access RenderDoc capture data and assist with graphics debugging.
Architecture
Claude/AI Client (stdio)
│
▼
MCP Server Process (Python + FastMCP 2.0)
│ File-based IPC (%TEMP%/renderdoc_mcp/)
▼
RenderDoc Process (Extension)Since RenderDoc's built-in Python does not include the socket module, communication is done via file-based IPC.
Related MCP server: RenderDoc MCP Server
Setup
1. Install the RenderDoc extension
python scripts/install_extension.pyThe extension is installed to %APPDATA%\qrenderdoc\extensions\renderdoc_mcp_bridge.
2. Enable the extension in RenderDoc
Launch RenderDoc
Tools > Manage Extensions
Enable "RenderDoc MCP Bridge"
3. Install the MCP server
uv tool install
uv tool update-shell # PATHに追加After restarting your shell, the renderdoc-mcp command becomes available.
Note: Adding
--editablereflects source code changes immediately (useful during development). To install as a stable version, useuv tool install ..
4. Configure the MCP client
Claude Desktop
Add to claude_desktop_config.json:
{
"mcpServers": {
"renderdoc": {
"command": "renderdoc-mcp"
}
}
}Claude Code
Add to .mcp.json:
{
"mcpServers": {
"renderdoc": {
"command": "renderdoc-mcp"
}
}
}Usage
Launch RenderDoc and open a capture file (.rdc)
Access RenderDoc data from an MCP client (such as Claude)
MCP tool list
Tool | Description |
| Check the capture loading status |
| Get the draw call list in a hierarchical structure |
| Get detailed information for a specific draw call |
| Get shader source code and constant buffer values |
| Get buffer contents (Base64) |
| Get texture metadata |
| Get texture pixel data (Base64) |
| Get pipeline state |
Usage examples
Getting the draw call list
get_draw_calls(include_children=true)Getting shader information
get_shader_info(event_id=123, stage="pixel")Getting pipeline state
get_pipeline_state(event_id=123)Getting texture data
# 2Dテクスチャのmip 0を取得
get_texture_data(resource_id="ResourceId::123")
# 特定のmipレベルを取得
get_texture_data(resource_id="ResourceId::123", mip=2)
# キューブマップの特定の面を取得 (0=X+, 1=X-, 2=Y+, 3=Y-, 4=Z+, 5=Z-)
get_texture_data(resource_id="ResourceId::456", slice=3)
# 3Dテクスチャの特定の深度スライスを取得
get_texture_data(resource_id="ResourceId::789", depth_slice=5)Getting partial buffer data
# バッファ全体を取得
get_buffer_contents(resource_id="ResourceId::123")
# オフセット256から512バイト取得
get_buffer_contents(resource_id="ResourceId::123", offset=256, length=512)Requirements
Python 3.10+
RenderDoc 1.20+
Note: Operation has only been verified on Windows + DirectX 11. It may also work on Linux/macOS + Vulkan/OpenGL environments, but this has not been verified.
License
MIT
Available Tools
15 toolsfind_draws_by_resourceA
Find all draw calls using a specific resource ID (exact match).
Args: resource_id: Resource ID to search for (e.g. "ResourceId::12345" or "12345")
Returns a list of matching draw calls with event IDs and match reasons. Searches shaders, SRVs, UAVs, render targets, and depth targets.
Inputs: resource_id (required). Returns: object
find_draws_by_shaderA
Find all draw calls using a shader with the given name (partial match).
Args: shader_name: Partial name to search for in shader names or entry points stage: Optional shader stage to search (if not specified, searches all stages)
Returns a list of matching draw calls with event IDs and match reasons.
Inputs: shader_name (required), stage. Returns: object
find_draws_by_textureA
Find all draw calls using a texture with the given name (partial match).
Args: texture_name: Partial name to search for in texture resource names
Returns a list of matching draw calls with event IDs and match reasons. Searches SRVs, UAVs, and render targets.
Inputs: texture_name (required). Returns: object
get_action_timingsA
Get GPU timing information for actions (draw calls, dispatches, etc.).
Args: event_ids: Optional list of specific event IDs to get timings for. If not specified, returns timings for all actions. marker_filter: Only include actions under markers containing this string (partial match). exclude_markers: Exclude actions under markers containing these strings.
Returns timing data including:
available: Whether GPU timing counters are supported
unit: Time unit (typically "seconds")
timings: List of {event_id, name, duration_seconds, duration_ms}
total_duration_ms: Sum of all durations
count: Number of timing entries
Note: GPU timing counters may not be available on all hardware/drivers.
Inputs: event_ids, marker_filter, exclude_markers. Returns: object
get_buffer_contentsA
Read the contents of a buffer resource.
Args: resource_id: The resource ID of the buffer to read offset: Byte offset to start reading from (default: 0) length: Number of bytes to read, 0 for entire buffer (default: 0)
Returns buffer data as base64-encoded bytes along with metadata.
Inputs: resource_id (required), length, offset. Returns: object
get_capture_statusA
Check if a capture is currently loaded in RenderDoc. Returns the capture status and API type if loaded.
Inputs: no named parameters. Returns: object
get_draw_call_detailsA
Get detailed information about a specific draw call.
Args: event_id: The event ID of the draw call to inspect
Includes vertex/index counts, resource outputs, and other metadata.
Inputs: event_id (required). Returns: object
get_draw_callsA
Get the list of all draw calls and actions in the current capture.
Args: include_children: Include child actions in the hierarchy (default: True) marker_filter: Only include actions under markers containing this string (partial match) exclude_markers: Exclude actions under markers containing these strings (list of partial matches) event_id_min: Only include actions with event_id >= this value event_id_max: Only include actions with event_id <= this value only_actions: If True, exclude marker actions (PushMarker/PopMarker/SetMarker) flags_filter: Only include actions with these flags (list of flag names, e.g. ["Drawcall", "Dispatch"])
Returns a hierarchical tree of actions including markers, draw calls, dispatches, and other GPU events.
Inputs: event_id_max, event_id_min, flags_filter, only_actions, +3 more. Returns: object
get_frame_summaryA
Get a summary of the current capture frame.
Returns statistics about the frame including:
API type (D3D11, D3D12, Vulkan, etc.)
Total action count
Statistics: draw calls, dispatches, clears, copies, presents, markers
Top-level markers with event IDs and child counts
Resource counts: textures, buffers
Inputs: no named parameters. Returns: object
get_pipeline_stateA
Get the full graphics pipeline state at a specific event.
Args: event_id: The event ID to get pipeline state at
Returns detailed pipeline state including:
Bound shaders with entry points for each stage
Shader resources (SRVs): textures and buffers with dimensions, format, slot, name
UAVs (RWTextures/RWBuffers): resource details with dimensions and format
Samplers: addressing modes, filter settings, LOD parameters
Constant buffers: slot, size, variable count
Render targets and depth target
Viewports and input assembly state
Inputs: event_id (required). Returns: object
get_shader_infoA
Get shader information for a specific stage at a given event.
Args: event_id: The event ID to inspect the shader at stage: The shader stage (vertex, hull, domain, geometry, pixel, compute)
Returns shader disassembly, constant buffer values, and resource bindings.
Inputs: stage (required), event_id (required). Returns: object
get_texture_dataA
Read the pixel data of a texture resource.
Args: resource_id: The resource ID of the texture to read mip: Mip level to retrieve (default: 0) slice: Array slice or cube face index (default: 0) For cube maps: 0=X+, 1=X-, 2=Y+, 3=Y-, 4=Z+, 5=Z- sample: MSAA sample index (default: 0) depth_slice: For 3D textures only, extract a specific depth slice (default: None = full volume) When specified, returns only the 2D slice at that depth index
Returns texture pixel data as base64-encoded bytes along with metadata including dimensions at the requested mip level and format information.
Inputs: resource_id (required), mip, slice, sample, +1 more. Returns: object
get_texture_infoC
Get metadata about a texture resource.
Args: resource_id: The resource ID of the texture
Includes dimensions, format, mip levels, and other properties.
Inputs: resource_id (required). Returns: object
list_capturesA
List all RenderDoc capture files (.rdc) in the specified directory.
Args: directory: The directory path to search for capture files
Returns a list of capture files with their metadata including:
filename: The capture file name
path: Full path to the file
size_bytes: File size in bytes
modified_time: Last modified timestamp (ISO format)
Inputs: directory (required). Returns: object
open_captureA
Open a RenderDoc capture file (.rdc).
Args: capture_path: Full path to the capture file to open
Returns success status and information about the opened capture. Note: This will close any currently open capture.
Inputs: capture_path (required). Returns: object
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 15 tools
Each tool targets a distinct concern: retrieval of specific resources (buffers, textures, pipeline state), listing or querying actions (draw calls, frame summary), searching by criteria (shader, texture, resource), and file management (list/open captures). There is no overlap that would cause an agent to pick the wrong tool.
Tool names follow a consistent verb_noun pattern: get_ for direct retrieval, find_draws_by_ for search operations, list_ and open_ for capture file management. The verbs clearly indicate the action type, and nouns specify the target, making the naming predictable and intuitive.
At 15 tools, this server is well-scoped for a graphics debugging domain. Each tool serves a meaningful analysis or management purpose, and the count is within the ideal range without redundancy or excessive granularity.
The tool surface covers the primary RenderDoc workflows: opening captures, inspecting frames and draw calls, retrieving resources, and searching by resource usage. Minor gaps exist (e.g., no tool to fetch rendered output images or detailed mesh data), but agents can accomplish core analysis tasks without dead ends.
Maintenance
Related MCP Connectors
Connect your AI assistant to Blender for 3D modeling, Geometry Nodes, materials, lighting, animation, rendering and export. Inspect real scenes and viewport images, refine the result and restore checkpoints. OAuth connects your chat to Blender Bridge Desktop on an online Windows 10/11 x64 computer running Blender 4.2+. The chat can run on another device. Requires an active Blender Bridge account; 7-day trial available. AI model access is separate.
Live browser debugging for AI assistants — DOM, console, network via MCP.
Search your AI chat history (ChatGPT, Claude, Codex) from any MCP client. Remote, private, read-only
Read, query, edit and render DWG drawings with AI. No AutoCAD required.
Related MCP Servers
- FlicenseNot gradedqualityDmaintenanceEnables AI assistants to analyze RenderDoc GPU frame captures for graphics debugging and performance analysis through a headless Python API. It provides a comprehensive suite of tools for inspecting pipeline states, shader bindings, and exporting frame resources like textures and meshes.183-
- AlicenseAqualityNot gradedmaintenanceEnables AI assistants to analyze and inspect RenderDoc capture files, providing tools to examine draw calls, textures, buffers, and shader information. It allows developers to perform graphics debugging and resource analysis through natural language interactions.12MIT
- AlicenseNot gradedqualityDmaintenanceExposes RenderDoc capture analysis to AI clients via MCP, enabling offline analysis of .rdc files (action tree, pipeline state, textures, shaders) and GUI state interaction on Windows.2MIT
- AlicenseAqualityDmaintenanceEnables AI assistants to access RenderDoc capture data and assist in graphics debugging via a file-based IPC bridge.15520MIT