itch.io

Thanks for checking! Both files are valid — .mcp.json is for project-scoped MCP servers (only loaded when you launch Claude Code from that project directory), while .claude.json in your home directory applies globally to all projects. Since Godot MCP Pro is typically used per-project, .mcp.json is actually the recommended scope. You can use .claude.json instead if you'd prefer it available everywhere, though.

Hey, thanks for trying it out!

Quick disclaimer up front: Godot MCP Pro was developed and tested primarily against Claude (Sonnet/Opus), so I can't guarantee identical reliability with local models. But here's what the current (May 2026) landscape looks like for your VRAM budget:

### Top pick for 24GB VRAM + MCP tool calling

**Qwen3.6-27B (dense)** — released April 2026, Apache 2.0, 262K context, native tool-calling. SWE-bench Verified 77.2. Fits comfortably in 24GB at Q4_K_M (~17–18GB) with room for context. This is currently the best single bet for agentic tool use at your VRAM tier.

### Strong alternatives

- **Qwen3-Coder-Next (80B MoE, 3B active)** — best-in-class for agentic coding (SWE-Bench-Pro at the level of models 10–20× larger), but it needs ~46GB VRAM/RAM at Q4. Won't fit 24GB GPU-only — workable only if you have unified memory (Apple Silicon) or are OK offloading to system RAM with a big speed hit.

- **Qwen2.5-Coder-32B-Instruct** — still excellent for FIM / tab-completion. Tool calling is decent but Qwen3.6-27B is a clearer upgrade for agent workflows.

- **Mistral Small 3.x (24B)** — solid function calling, fast, fits comfortably. A reasonable middle option if Qwen feels too verbose.

### What I'd avoid (or be cautious with)

- **DeepSeek V3** — independent evaluations have it around 81.5% on function-calling benchmarks vs ~96.5% for Qwen Plus, and it's specifically weaker at multi-turn function calling. For single-shot ("create this scene") it's fine; for chained Godot workflows (create scene → add nodes → write script → reload → test), expect drift.

- **Llama 3.x at this VRAM tier** — 70B doesn't fit cleanly at 24GB even quantized, and 8B isn't reliable enough for multi-step tool chains.

### What matters more than the model name

1. **Use a runtime that enforces tool schemas.** llama.cpp grammars, vLLM with guided decoding, or Ollama with native tool support. Free-form decoding will hallucinate tool args no matter which model you pick.

2. **Multi-step Godot workflows are the hard part.** Anything past 3–4 chained tool calls (which is most non-trivial editor work) is where local models drop off vs Claude. Expect to babysit longer chains.

3. **Keep the system prompt tight.** Local models burn context fast, and Godot MCP Pro's tool surface is broad — consider trimming the available tool list if you only need a subset.

If you do find a setup that works smoothly, I'd love to hear about it — drop a note here or on Discord.

Hi Jdavz — really sorry, this is a packaging bug on our side, not anything you did. The ZIP we shipped had a nested folder layout that Blender's "Install from Disk" couldn't recognize, which is why the addon never appeared in your add-ons list. I just pushed **v1.3.2** which fixes it.

**What to do:**

1. Re-download the latest ZIP from your itch.io download page — it should be `blender-mcp-pro-v1.3.2-itch.zip`.

2. Extract it. Inside you'll see:

   - `blender_mcp_pro.zip` ← **this is the addon**

   - `server/`

   - `README.md`

3. In Blender 4.2.2: **Edit > Preferences > Add-ons** → click the dropdown (▾) at the top right → **Install from Disk…** → select **`blender_mcp_pro.zip`** (the inner one, NOT the outer download).

4. Tick the checkbox next to "Blender MCP Pro" to enable it.

5. Open the 3D Viewport sidebar (press **N**) → **BM Pro** tab → **Start Server**.

After that, on the MCP server side: `pip install -r server/requirements.txt` and add the server to your MCP client config as in the README. Happy to walk you through that step too once the addon is enabled — just reply here with which client you're using (Claude Desktop / Claude Code / Cursor / Windsurf).

Thanks for your patience and for purchasing — this should have just worked.

Thanks for the report — you were right about the parameter-name mismatch between the server schemas and the plugin. v1.0.2 is now up with this fixed (widget_path / blueprint_path and the rest are aligned).

After downloading, re-run:

```

cd server

npm install

npm run build

```

Sorry for the trouble, and thanks for the clear repro.

— Youichi

Thank you!!!

We don't have a formal roadmap, but we regularly get feature requests from users on Discord and other channels, so we pick those up and work on them as they come in!

Just released v1.10.2 with this fix included. Thanks again for reporting it!

Thank you so much for sharing this, and for the kind words! Really glad it's working well for you.

And sorry for the late response to your email — I should have addressed the permission issue sooner.

The OpenCode configuration example is super helpful for other users — thanks for posting it here too.

Regarding the permission issue on Linux, you're right — we'll add a chmod step to the build process in the next release so it works out of the box. For now, your workaround is exactly right.

Thanks again for the feedback!

I looked into adding Alipay before, but it's pretty difficult to set up for non-Chinese sellers. Both itch.io and Buy Me a Coffee use Stripe/PayPal which don't support Alipay for my region.

Can you use PayPal? itch.io supports PayPal checkout which might be easier for you. Visa/Mastercard also works on both platforms. Otherwise let me know and we can figure something out!

Thanks for sharing your experience — this is really helpful feedback!

You're right that the local LLM setup is much harder than it should be, especially for non-programmers. The sticking/freezing issue with Ollama + Roo Code is a known pain point — local models often struggle with the large number of tool definitions and can timeout or get stuck in loops.

A step-by-step video tutorial for local setup is a great idea and I'll put it on my to-do list. In the meantime, a couple of tips if you want to try again:

- Use `--minimal` mode (35 tools instead of 169) — this drastically reduces the context size that chokes local models

- Gemma4 26B should work with `--minimal`, but give it simpler one-step instructions rather than complex multi-part requests

- If Ollama freezes, it's usually the model running out of context window, not the MCP plugin

Glad you joined the Discord — drop a message there anytime and I'll help you get it running!

Great questions!

1. Yes, it works 100% locally with no cloud or subscription needed. The MCP server and Godot plugin run entirely on your machine. However, you still need an AI client to send the commands — Godot MCP Pro is the bridge between the AI and the editor, not the AI itself. For local LLMs like Gemma4, you'd use something like LM Studio or Ollama + OpenCode/Continue as the AI client. With an RTX 5090 you have plenty of power for that. Use `--minimal` mode (35 tools) to keep the context small for local models.

2. No, there's no chat panel inside Godot. You chat with the AI in your AI client (Claude Code, Cursor, VS Code Copilot, etc.) and it controls Godot through the MCP connection. Think of it as: you talk to the AI in one window, and it makes changes in the Godot editor in real time.

3. Yes, that kind of natural language prompt absolutely works! The AI translates your description into the actual Godot operations. Your laser example would work — the AI would create a scene, set up the icon.png as a sprite, write the WASD movement script, add the shooting mechanic with laser.png, and wire up spacebar input. You don't need to know GDScript — that's the whole point.

That said, the quality of results depends on the AI model. Larger cloud models (Claude, GPT) will understand your intent better than smaller local models. With Gemma4 26B/31B you might need to be a bit more specific in your descriptions, but it should still work for basic game mechanics.

Feel free to join our Discord if you have more questions or need help setting up: https://discord.gg/zJ2u5zNUBZ

Great question! Version numbers follow semver, so 1.9 doesn't mean we're close to 2.0 — it'll keep going 1.10, 1.11, 1.20, etc. I don't have any plans for a v2 right now, there's still plenty of room to grow in v1.x. So your purchase covers a lot of future updates!

Just shipped it! v1.1.0 adds streamable HTTP transport for Aseprite MCP Pro.

Start the server with:

  node server/build/index.js --http

Then configure your client:

  {

    "mcpServers": {

      "aseprite-mcp-pro": {

        "type": "streamable-http",

        "url": "http://127.0.0.1:8003/mcp"

      }

    }

  }

Port is customizable with --http-port or the ASEPRITE_MCP_HTTP_PORT env var (default: 8003).

Let me know how it goes!

Quick update: v1.3.0 just shipped with streamable HTTP transport support!

You can now run the server in HTTP mode:

python server/server.py --transport streamable-http --port 8000

Default is still stdio (no changes needed for existing setups). The HTTP mode enables remote connections and web-based MCP clients.

Thanks for the suggestion @InfernalDread!

Thanks for the questions!

License verification: The itch.io version has no license verification at all — it runs entirely offline. The license check only exists in the Gumroad subscription version (to manage monthly billing). So with the itch.io purchase, there's nothing to revoke. It's yours.

venv: Yes, absolutely. A venv is the recommended approach actually. Create one, install requirements.txt into it, then point your MCP client config to the venv's Python path (e.g. "command": "blender-mcp-env/Scripts/python").

Streamable HTTP: Currently the MCP server uses stdio transport (the standard for local MCP servers). Adding streamable HTTP transport is an interesting idea — it would allow remote connections and web-based clients. I'll look into it for a future update. Thanks for the suggestion!

Thanks for the feedback! I've already applied this fix on our end — print() → print_verbose() for the connect/disconnect messages, exactly the same change your Claude made.

I just released v1.8.1 though, so this will be included in the next update. Sorry for the wait, but hang tight until then!

Sorry. The correct URL is this.
https://discord.gg/N6sSHBGuYX

Hi it's there

https://github.com/youichi-uda/godot-mcp-pro/blob/master/CHANGELOG.md

Hey, no bother at all — great suggestion!

I've added Streamable HTTP/SSE transport support for all three servers (Unreal, Godot, and Unity). This will let you start the server with an `--http` flag and connect your MCP client to something like `http://127.0.0.1:8000/mcp` — much better for remote workflows.

This will be included in the next release for each server. Thanks for the feedback!

Thank you for reporting this! You're absolutely right — the server folder was accidentally missing from the v1.0.0 zip. I sincerely apologize for the inconvenience.

I've just released v1.0.1 which includes both the `plugin/` and `server/` folders. Please re-download and you should be all set.

After downloading, run the following in the `server/` directory to get started:

```

npm install

npm run build

```

Again, sorry about that, and thanks for letting me know!

Thank you for the encouraging feedback! It's great to know you're enjoying it with LM Studio. We're fully committed to continuing support and development — comments like yours make all the difference. Happy game making! 🎮

Thank you for letting me know! My invitation code had expired. This time I made one that doesn't expire.

https://discord.gg/zJ2u5zNUBZ

Yes, it does! If anything comes up, feel free to reach out on Discord — happy to help!

Yes!!

Great to hear! Mac is now a supported platform.

No compiled binaries — it's all GDScript (Godot plugin) and TypeScript source (MCP server), so it should work cross-platform. That said, I've only tested on Windows so far.

If you're on Mac and willing to help verify compatibility, I'd be happy to give you a free copy in exchange for testing feedback. Just reach out:

- Discord: https://discord.gg/F4gR739y

- Email: abyo.software@gmail.com