Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/davidbuenov/dbv-mcp-server/llms.txt

Use this file to discover all available pages before exploring further.

Claude Desktop communicates with MCP servers over stdio transport: it launches a process, writes JSON-RPC 2.0 messages to its stdin, and reads responses from its stdout. bridge.py acts as the translation layer between that stdio channel and the HTTP-SSE protocol that Unreal Engine’s built-in MCP server uses on port 8000.

​
How it works

Claude Desktop (stdio)
       │  JSON-RPC 2.0 on stdin/stdout
       ▼
  bridge.py
  ┌──────────────────────────────────┐
  │  tools/list  → flattens toolsets │
  │  tools/call  → namespace rewrite │
  │  debug logs  → stderr only       │
  └──────────────────┬───────────────┘
                     │ HTTP POST + SSE
                     ▼
      Unreal Engine MCP Server (:8000)
bridge.py reads JSON-RPC messages line-by-line from stdin, proxies them as HTTP POST requests to http://localhost:8000/mcp, reads the SSE response stream, and writes compact single-line JSON back to stdout — all without polluting the JSON-RPC channel with log output (logs go to stderr).

​
Prerequisites

  • Unreal Engine 5.8 with the ModelContextProtocol plugin enabled and the MCP server running on port 8000.
  • Python 3.x installed on the machine running Claude Desktop.
  • The dbv-mcp-server repository cloned locally.
  • No additional Python packages are required — bridge.py uses only the standard library.

​
Configuration

Edit claude_desktop_config.json to register bridge.py as an MCP server. Claude Desktop reads this file at startup and launches each entry as a child process. 
{
  "mcpServers": {
    "unreal-engine-mcp": {
      "command": "C:\\Program Files\\Python312\\python.exe",
      "args": [
        "-u",
        "C:\\ABSOLUTE\\PATH\\TO\\dbv-mcp-server\\bridge.py"
      ]
    }
  }
}
Windows: always use the full absolute path to python.exe. Claude Desktop may not inherit your user PATH, so python or python3 can silently fail to launch. Find the correct path with where python in a terminal. Also use double-escaped backslashes (\\) in JSON strings — a single \ is a JSON escape character.

​
Where to find claude_desktop_config.json

PlatformLocation
macOS~/Library/Application Support/Claude/claude_desktop_config.json
Windows (standard installer)%APPDATA%\Claude\claude_desktop_config.json
Windows (MSIX / sandbox install)%LOCALAPPDATA%\Packages\Claude_pzs8sxrjxfjjc\LocalCache\Roaming\Claude\claude_desktop_config.json
After editing the file, fully quit and restart Claude Desktop — changes are only picked up at startup.

​
Available tools after connecting

Once Claude Desktop launches bridge.py and completes the MCP handshake, it will see a flat list of tools that covers:
ToolDescription
list_toolsetsList all active toolsets with names and descriptions
describe_toolsetGet all tool names, descriptions, and input schemas for a given toolset
call_toolCall any toolset tool by name and pass arguments
EditorToolset toolsViewport, camera, selection, assets, Blueprint authoring, material editing, scene spawning, actor management, and more (+100 tools)
AutomationTestToolset toolsDiscoverTests, ListTests, RunTests, GetResults
SlateInspectorToolset toolsClick, Type, Snapshot, Observe, Windows, and more
The flattening is done automatically by bridge.py at connection time. The first tools/list call triggers a live query to Unreal — subsequent calls are served from the on-disk cache (.mcp_tools_cache.json) until the toolset list changes.
For AI-assisted Python scripting (RunPython), register the companion dbv-unreal-python-api MCP server alongside bridge.py. This gives Claude access to both the live Unreal toolset and the full 11,600-class Python API reference: 
{
  "mcpServers": {
    "unreal-engine-mcp": {
      "command": "C:\\Program Files\\Python312\\python.exe",
      "args": ["-u", "C:\\ABSOLUTE\\PATH\\TO\\dbv-mcp-server\\bridge.py"]
    },
    "dbv-unreal-python-api": {
      "command": "C:\\Program Files\\Python312\\python.exe",
      "args": [
        "-u",
        "C:\\ABSOLUTE\\PATH\\TO\\dbv-mcp-server\\skills\\dbv-unreal-python-api\\scripts\\mcp_server.py"
      ]
    }
  }
}
See the Python API Skill docs for the full setup guide and macOS/Linux config.

​
Custom Instructions: make Claude always consult the API

By default, Claude will consult the Python API skill only when it judges the query to be relevant — the MCP protocol offers no way to force tool use. To substantially increase the likelihood that Claude always searches the API before writing RunPython scripts, add a persistent instruction in Claude Desktop → Settings → Custom Instructions: 
When writing Python scripts for Unreal Engine (RunPython), always:
1. Call dbv_unreal_dev_guide first to understand the conceptual workflow.
2. Call dbv_python_unreal_api to verify exact class/method signatures.
3. Never assume API names that were not returned by those tools.
This mirrors the automatic enforcement that Claude Code provides via the .claude/skills/dbv-unreal-python-api/SKILL.md project skill. In Claude Desktop the instruction is advisory rather than enforced, but in practice it works well for standard development workflows.

Build docs developers (and LLMs) love

Get started for freeTalk to us