Provides screenshot and display capture functionality using AppleScript, enabling agents to visually monitor and evaluate Godot game development tasks.
Click on "Install 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., "@GameDevBench MCP capture a screenshot of the current Godot game window"
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.
A benchmark suite for evaluating LLM agents on game development tasks.
Paper: GameDevBench: A Comprehensive Benchmark for Game Development
GameDevBench contains 132 game development tasks to evaluate LLM agents' ability to complete game development problems in the Godot game engine.
Related MCP server: PeepIt MCP
Godot 4.x - Download and install from godotengine.org
Ensure godot is available in your PATH, or set GODOT_EXEC_PATH environment variable
Python 3.10+ - Required for all agents
Python 3.12+ - Required for OpenHands agent
Install the agent(s) you want to use:
Claude Code - Claude Code
Codex - Codex
Gemini CLI - Gemini CLI
OpenHands - OpenHands
Before running the benchmark, unzip the tasks folder:
unzip tasks.zipNote: The tasks are distributed as a zip file to prevent accidental data leakage.
You can use the built-in plans for claude-code, codex, and gemini-cli, or provide API keys directly. For OpenHands you must provide your own API keys. See .env.example for a complete list of optional environment variables.
uv run python gamedevbench/src/benchmark_runner.py \
--agent AGENT \
--model MODEL \
run --task-list tasks.yamlclaude-code - Anthropic's Claude Code CLI
codex - OpenAI Codex
gemini-cli - Google Gemini CLI
openhands - OpenHands (requires Python 3.12+)
--agent AGENT - Agent to use (required)
--model MODEL - Model name (e.g., claude-sonnet-4.5-20250929)
--enable-mcp - Enable MCP (Model Context Protocol) server for supported agents
Provides screenshot capabilities to the agent
Note: MCP server requires macOS (see limitations below)
--use-runtime-video - Enable runtime video mode
Appends Godot runtime instructions to prompts
Helps agents understand how to run and test their changes
--skip-display - Skip tasks that require display
run --task-list FILE - Run tasks from YAML file (e.g., tasks.yaml)
macOS-only Features:
MCP server screenshot functionality (--enable-mcp) currently only works on macOS
Uses AppleScript for display capture
Requires setting GODOT_SCREENSHOT_DISPLAY environment variable to correct display number
Benchmark results are saved to results/ directory with the following information:
Task success/failure status
Token usage and costs
Execution time
Validation results
@misc{chi2026gamedevbenchevaluatingagenticcapabilities,
title={GameDevBench: Evaluating Agentic Capabilities Through Game Development},
author={Wayne Chi and Yixiong Fang and Arnav Yayavaram and Siddharth Yayavaram and Seth Karten and Qiuhong Anna Wei and Runkun Chen and Alexander Wang and Valerie Chen and Ameet Talwalkar and Chris Donahue},
year={2026},
eprint={2602.11103},
archivePrefix={arXiv},
primaryClass={cs.AI},
url={https://arxiv.org/abs/2602.11103},
}Launch Godot editor on specified display and take a screenshot
| Name | Required | Description | Default |
|---|---|---|---|
| project_dir | Yes | Path to the Godot project directory | |
| display | No | No need to specify this, it will be automatically determined by the MCP server. Do not specify this if you are not sure about the display number. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations provided, the description carries the full burden of behavioral disclosure. While it mentions launching the editor (a side effect), it fails to disclose critical behavioral traits: whether the editor process remains open after screenshotting, the output format/destination of the screenshot, cleanup behavior, or required environment setup.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single, efficient sentence with no redundant words. The action verbs are front-loaded, and every word contributes to understanding the tool's function.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the complexity (external process launch, GUI automation, image capture) and absence of output schema or annotations, the description is incomplete. It fails to specify the return value format (critical for an agent to use the screenshot) or side effect persistence (editor state).
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100%, establishing a baseline of 3. The description mentions 'specified display' (referencing the display parameter) and implies project_dir is used to launch that project, but adds no semantic information about valid formats, constraints, or the warning (from schema) that display should typically be omitted.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the specific actions: 'Launch Godot editor' and 'take a screenshot'. However, it omits what happens to the screenshot (returned as data? saved to file?) and doesn't distinguish scope since there are no sibling tools.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description provides no guidance on when to use this versus alternatives, prerequisites (e.g., Godot installation), or when not to use it. The phrase 'specified display' contradicts the parameter schema which advises against specifying the display.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections. Dates show when Glama detected each change.
v0.1.0godot-screenshotWith only one tool, there is no possibility of ambiguity or overlap between tools. The tool's purpose is clearly defined and distinct by default.
A single tool inherently has perfect naming consistency, as there are no other tools to compare it against. The name 'godot-screenshot' follows a clear noun-verb pattern.
A single tool for a server named 'GameDevBench MCP' suggests an extremely limited scope that does not align with the implied breadth of game development tasks. This feels thin and incomplete for the domain.
The tool set is severely incomplete for game development, covering only a niche screenshot function. There are obvious gaps in core operations like asset management, testing, or build processes, making it inadequate for the stated purpose.
Unclaimed servers have limited discoverability.
If you are the server author, to access and configure the admin panel.
Generate images, GIFs, and PDFs from HTML, URLs, or templates — from your AI agent.
OCR, transcription, file extraction, and image generation for AI agents via MCP.
Operate Linux, macOS and Windows from your LLM. Every action runs through an auditable allowlist.
Control real Android and iOS devices with LLM agents — tap, swipe, type, automate flows.
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/SeeleAI/gamedevbench-mcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server
