Game particle effects in Claude Code

Claude Code already has your game open: it reads the scenes, writes the scripts and runs the build. With the ParticleFX MCP server it can make the effects as well. It starts from one of the library effects or a blank canvas, edits it, looks at real renders of its own work, and exports a spritesheet with the data file your engine reads, then moves the files into place itself.

Connect Claude Code

1. Sign in at particlefx.studio and make an API key on your billing page. Any account can, the free one included. The key starts with pfx_live_, is shown once, and can be revoked there at any time.
2. Add the server from a terminal in your project. Without --scope it is added for you, in this project only; --scope user makes it available in every project.
3. Run claude mcp list, or /mcp inside a session: particlefx should show as connected, with its tools.

export PFX_API_KEY=pfx_live_your_key
claude mcp add --transport http particlefx https://mcp.particlefx.studio/mcp \
  --header "Authorization: Bearer $PFX_API_KEY"

To share the setup with your team without sharing a key, add it with --scope project. That writes .mcp.json in the project root, which you can commit; Claude Code fills in ${PFX_API_KEY} from each person's environment, and each person approves the server once.

{
  "mcpServers": {
    "particlefx": {
      "type": "http",
      "url": "https://mcp.particlefx.studio/mcp",
      "headers": { "Authorization": "Bearer ${PFX_API_KEY}" }
    }
  }
}

You can also skip the key: add the server without the --header line, then run /mcp in a session (or claude mcp login particlefx in a terminal) and authenticate. Claude Code opens particlefx.studio, you sign in and allow the connection, and calls are billed to that account. Use one or the other: with a header set, a rejected key shows as a failed connection rather than offering sign-in.

Claude Code asks before it runs a tool it has no permission rule for. To stop it asking for each ParticleFX tool, answer "Yes, don't ask again", or add the permission rule mcp__particlefx, which covers every tool of the server.

Example: a looping fireball for Godot 4

Ask in your own words. This prompt names the effect and the folder, so the files land where your scenes expect them:

Use the particlefx MCP server. Start from the Pixel Fireball library effect and name it
fireball. Make the trail a bit longer, check the render, then export it as a spritesheet
with Godot 4 SpriteFrames data, transparent background, 30 fps. Put fireball.png and
fireball.tres in res://fx/ and fix the texture path in the .tres.

What happens, tool by tool:

1. list_presets finds Pixel Fireball (pixel-arcade/fireball) in the library.
2. create_effect copies it into your account as fireball. From here on it is your effect: it shows up on My effects and open_in_studio gives a link to it.
3. edit_effect applies the change you asked for as edit ops, the same edit language the studio and its own agent use.
4. render_preview returns a labelled contact sheet of moments across the loop. Claude Code looks at it, and edits again if something reads badly. analyze_effect checks that the loop closes and counts particles over it.
5. export_effect with format: "sheet", sheet_meta: "godot", transparent: true and fps: 30 renders the real thing at full size and returns a download link that works for 1 hour.
6. Claude Code downloads the zip and unpacks it with curl and unzip, into fx/. The zip holds fireball.png and fireball.tres.

The .tres is a SpriteFrames resource with one animation named fireball, looping at the exported 30 fps. It points at res://fireball.png, which is why the prompt asks for the path fix; after it, the texture line reads:

[ext_resource type="Texture2D" path="res://fx/fireball.png" id="1_tex"]

Playing it is a few lines. Claude Code can write this part too:

extends Node2D

const FIREBALL := preload("res://fx/fireball.tres")

func spawn_fireball(at: Vector2) -> AnimatedSprite2D:
	var fx := AnimatedSprite2D.new()
	fx.sprite_frames = FIREBALL
	fx.position = at
	add_child(fx)
	fx.play("fireball")
	return fx

• Pixel art: set Texture Filter to Nearest on the node, or project-wide under Rendering, Textures.
• One-shot effects (a hit, an explosion): call FIREBALL.set_animation_loop("fireball", false) once and connect animation_finished to queue_free.
• More on the resource and the free route without it: Godot 4 engine guide.

Pixel FireballWarp PortalHealWall Torch

What it costs

Connecting costs nothing, and neither do searching the library, reading the reference, analysing an effect, or any edit after the first. Paid steps use builds, the same builds as on the website:

• Create the effect from a library preset (create_effect): 1 build.
• Edits and the loop check (edit_effect, analyze_effect): free.
• Previews (render_preview): the first 20 of each UTC day free, then 1 build per 4, at most 200 a day.
• Export (export_effect, Godot SpriteFrames): 1 build, and the Godot data file needs a paid plan.

So this example takes 2 builds of work if the previews stay inside the free ones: 1 for the new effect and 1 for the export. The free sign-up and daily builds are for the website and never pay for MCP calls. On a free account and on Starter, MCP calls are paid from pack builds, from $4.99 for 150 that never expire. Creator at +50%, Studio at +25%, Studio Max one for one can spend plan builds instead, and fall back to pack builds when those run out. Work that fails on our side is refunded.

The Godot data file is a plan feature, as in the studio: any plan, from $7.99 a month (Starter), unlocks it. Without a plan the same build exports a GIF, WebP, APNG, PNG frames, or the sheet as an image only (sheet_meta: "none").

If you would rather our own effects agent did the creative work, build_effect takes a description and costs 1 build (Basic) or 10 (Pro). Full price list: MCP tools and keys, limits and billing.