CVE-2026-25546: godot-mcp has Command Injection via unsanitized projectPath
godot-mcp (npm package @coding-solo/godot-mcp), an MCP server that lets AI agents drive the Godot game engine, passes the projectPath argument to a shell. Shell metacharacters in it run arbitrary commands with the privileges of the MCP server process. Versions below 0.1.1 are affected; 0.1.1 fixes it, and there is no workaround.
Published Feb 4, 2026 · Updated Feb 8, 2026 · Analysis Nov 1, 2026
What an attacker can do
The executeOperation function passed user-controlled input, such as projectPath, directly to exec(), which starts a shell. A value containing $(command), or a character like & that chains a command, runs that command. Every tool that takes projectPath is affected, including create_scene, add_node and load_sprite.
The argument is supplied by the AI agent calling the tool, not typed by a person. An attacker therefore does not need a direct connection to the server. They need to influence what the agent sends, for example through prompt injection in a repository, README, issue or web page the agent reads. The CVSS vector reflects this: local attack vector, user interaction required, no privileges.
The injected command runs as the user who started the MCP server. On a developer laptop, that gives access to source code, SSH keys, cloud credentials and tokens in the environment.
Who is affected
@coding-solo/godot-mcp below 0.1.1. Anyone who has registered this server with an MCP client such as an editor or coding agent is affected, whether or not they work on Godot projects that day, because the vulnerable tools are always exposed. Check the version:
npm ls -g @coding-solo/godot-mcpIf you run it through npx, look at the version pinned in your MCP client configuration.
@coding-solo/godot-mcp
- Ecosystem
- npm
- Affected
- < 0.1.1
- Fixed in
- 0.1.1
- Check
npm ls @coding-solo/godot-mcp
How to fix it
Upgrade to @coding-solo/godot-mcp 0.1.1. The fix switches from exec() to execFile(), which does not invoke a shell, so metacharacters in projectPath are no longer interpreted.
npm install -g @coding-solo/[email protected]The advisory lists no workaround. If you cannot upgrade, remove the server from your MCP client configuration until you can.
How to tell if you were hit
Look in shell history, process listings and audit logs for child processes spawned by the MCP server's Node process that are not Godot, such as curl, sh with unusual arguments, or anything run from a projectPath value. Check MCP client logs for tool calls such as create_scene or add_node whose projectPath contains $(, backticks, & or ;. Beam records agent shell commands and MCP tool calls, and its rules can flag a tool call with shell metacharacters in an argument, which helps trace what the agent was told to do.
Original advisory
From the GitHub Advisory Database (CC BY 4.0).
Impact
A Command Injection vulnerability in godot-mcp allows remote code execution. The executeOperation function passed user-controlled input (e.g., projectPath) directly to exec(), which spawns a shell. An attacker could inject shell metacharacters like $(command) or &calc to execute arbitrary commands with the privileges of the MCP server process.
This affects any tool that accepts projectPath, including create_scene, add_node, load_sprite, and others.
Patches
Fixed in version 0.1.1 by switching from exec() to execFile(), which does not invoke a shell.
Workarounds
None. Users should upgrade immediately.
Resources
- https://github.com/Coding-Solo/godot-mcp/issues/64
- https://github.com/Coding-Solo/godot-mcp/pull/67
Credited to TianYu-0829, wcole3, Coding-Solo.
Questions
- Is CVE-2026-25546 remote code execution?
The advisory calls it that. The vector is local with user interaction. The malicious value reaches the server through the AI agent, so the attacker needs to steer the agent, for example with prompt injection.
- Is there a workaround for godot-mcp?
No. The advisory states there is none and tells users to upgrade to 0.1.1 immediately.
- Which tools are affected?
Any tool that accepts projectPath, including create_scene, add_node and load_sprite.
References
How this page is made: the data panel comes from the GitHub Advisory Database, CISA’s KEV catalog, the NVD and FIRST EPSS, and updates twice a day. The analysis is written by an AI model from the sources listed above and checked by automated rules before it is published. Check versions against the vendor's advisory before you act. Methodology
More Other MCP servers and tools vulnerabilities
All Other MCP servers and tools entriesCVE-2026-102943
Dagu MCP DAG specification read bypasses path containment and workspace isolation
medium6.5Nov 10, 2026CVE-2026-102610
fast-agent: `fast-agent serve` binds HTTP to 0.0.0.0 with no authentication by default; with --shell this is unauthenticated remote command execution
medium6.4Nov 9, 2026CVE-2026-63655
tsParticles: Unbounded HTTP Request Body Buffering in @tsparticles/mcp-server Leads to Remote DoS
high7.5Nov 9, 2026CVE-2026-76892
tsParticles: Malformed JSON requests can terminate the @tsparticles/mcp-server HTTP process
high8.7Nov 9, 2026CVE-2026-76893
tsParticles: A single anonymous MCP session can monopolize HTTP mode and deny new clients
medium6.9Nov 9, 2026CVE-2026-77604
Code Graph RAG: MCP paginated read_file path traversal / arbitrary file read
high7.5Nov 9, 2026
