NVIDIA App Updates Ship with Default-Enabled Local Model Context Protocol Server
NVIDIA App 11.0.9.565 quietly enables a local Model Context Protocol (MCP) server on loopback port 13508. It exposes 12 system tools, half of which perform mutating host actions without per-call confirmations.
Why it matters
Check your NVIDIA App settings today to disable or isolate tool execution before unvetted local agents obtain app-launching and setting-mutation capabilities.
TL;DR
- 01NVIDIA App 11.0.9.565 runs an MCP server on loopback port 13508 with default-on permissions.
- 02The server exposes 6 mutating tools, including application launching and display capture.
- 03MCP protocol limits enforce binary access: all tools are permitted once connected without per-call confirmations.
Key facts
- Release version
- NVIDIA App 11.0.9.565
- Host binding
- 127.0.0.1:13508
- Tool endpoints
- 12 total (6 read-only, 6 mutating)
- Authorization type
- Binary toggle (all-or-nothing)
Default-On Local Surface
NVIDIA App version 11.0.9.565 introduces an embedded Model Context Protocol (MCP) server under the NvContainer process, bound to 127.0.0.1:13508. Following the update, the configuration flag Allow AI agents to control NVIDIA App defaults to active without explicit user prompts during installation.
Tool Capabilities: Read vs Mutate
The server implements 12 distinct tools divided evenly across query and execution primitives:
- Read-only (6): overlay status, current game info, installed applications, driver status, driver release notes, and laptop feature detection.
- Mutating (6):
nvapp_overlay_capture, display filter adjustments, telemetry configuration,nvapp_client_launch_application, game optimization application, and laptop power limits (Battery Boost and Whisper Mode).
Protocol-Level Security Gaps
Because the Model Context Protocol relies on a single binary connection gate, connecting any desktop agent grants immediate broad execution rights across all 12 endpoints. Although NVIDIA returns access_level_restricted in certain contexts, there is no transport-level requirement for per-call user confirmations. If you do not actively automate GPU profiles via local agents, navigate to Settings > Features > Connect Agents & Apps and toggle the server off.
Try it in 2 minutes
# Check if the NVIDIA MCP server port is currently listening
netstat -ano | findstr 13508
# Or inspect the local loopback via PowerShell
Test-NetConnection -ComputerName 127.0.0.1 -Port 13508bash
✓ When to use
- Use when actively writing automation scripts to adjust GPU clock limits, fan curves, or ShadowPlay recordings via local LLM tooling.
- Use behind a dedicated security policy firewall that audits every individual tool invocation.
✕ When NOT to use
- Do not leave enabled on multi-user developer workstations or machines running untrusted third-party agents.
- Do not route through reverse tunnels like ngrok without an upstream authentication proxy.
What to do today
- Open NVIDIA App, navigate to Settings > Features > Connect Agents & Apps, and verify the toggle state.
- Disable the toggle if you do not orchestrate GPU settings via trusted local coding agents.
- Place local MCP proxies in front of mutating agents to filter application execution payloads.