Steamworks MCP Server

steamworks-mcp

Put your game on Steam with your AI assistant. It reads your game project, asks you what it cannot find, writes your store page with you, checks everything against Valve's rules, and fills in Steamworks only after you say yes. It never publishes anything: you press Publish.

Works with Claude (Desktop, Code and claude.ai), ChatGPT (web and desktop app) and Codex (app, CLI and IDE extension). Free and open source (MIT).

37 tools · 36 skills · 24 code rules · 124 release checks · 114 store-text rules · Unity, Godot and Unreal

Not affiliated with or endorsed by Valve. "Steam" and "Steamworks" are trademarks of Valve Corporation.

Contents: How it works · Install · Your first conversation · Words you will see · What it can do · Skills · Code rules · Safety · Known limits · Advanced setup

How it works

flowchart LR
    A["Your game folder"] --> B["Scan: what the project<br/>already says"]
    B --> C["Questions for<br/>what is missing"]
    C --> D["Drafts: store text, achievements,<br/>images, settings, SDK code"]
    D --> E{"You approve?"}
    E -- "change it" --> D
    E -- "yes" --> F["Checks against<br/>Valve's rules"]
    F --> G["Files + a checklist<br/>for Steamworks"]
    F --> H["Fills Steamworks<br/>after a dry run and your OK"]
    G --> I["You press Publish<br/>in Steamworks"]
    H --> I

Your game goes through four release steps, the same ones Valve reviews. The tool shows how far each one is:

flowchart LR
    S0["Step 0 · Prerequisites<br/>account, bank, tax, app fee"] --> S1["Step 1 · Store page<br/>texts, images, tags, languages"]
    S1 --> S2["Step 2 · Build review<br/>the game, Steam features"]
    S2 --> S3["Step 3 · Release<br/>date, price, Release button"]

Everything about your game's Steam release lives in one readable file in your game folder, steamworks.yaml. Every value in it has a status: missing, draft (found or written, not yet checked by you), approved (you said it is right), or done in Steamworks.

Install

You need uv, a small tool that runs Python programs (one command to install), and Claude, ChatGPT or Codex. You do not need Steam keys or a Steam login to start.

The server speaks both MCP transports:

Over HTTP every request needs sign-in (OAuth, or a bearer token); over stdio only the app that started the server can talk to it.

Claude Desktop

In a terminal:

uvx steamworks-mcp setup

It asks where your games are and connects Claude Desktop (and Claude Code, if you have it). Restart Claude Desktop. To use the skills there too, add the folders in skills/ as skills in Claude's settings.

Claude Code: the plugin (server and skills in one step)

In Claude Code, type:

/plugin marketplace add https://github.com/wazzapsenk/steamworks-mcp
/plugin install steamworks@steamworks-mcp

It asks for your games folder. That's it.

ChatGPT (web or desktop app) and claude.ai

These apps reach the server over the internet, through a secure address on your computer.

1. Install cloudflared once (free, from Cloudflare). On Windows: winget install --id Cloudflare.cloudflared; on a Mac: brew install cloudflared.

2. In a terminal:

   uvx steamworks-mcp remote
   

   The first time, it asks where your games are. Then it prints an address that ends in /mcp and an access code.

3. Add the address in your app, with OAuth as the authentication:

   • ChatGPT on the web: Settings > Apps & Connectors > Advanced settings: turn on Developer mode, then create a connector.
   • ChatGPT desktop app (Windows, macOS): Settings > Plugins > MCPs > Add.
   • claude.ai: Settings > Connectors > Add custom connector.

4. A steamworks-mcp page opens and asks for the access code: type the one from step 2.

5. Ask your assistant: "Where are my games on Steam?"

Keep the terminal from step 2 open while you work. Custom connectors in ChatGPT need developer mode; check that your plan has it. The address changes every time you start remote (update the connector's URL then); a fixed address needs a little more setup.

Codex (app, CLI and IDE extension)

1. In a terminal (the same command as for Claude Desktop):

   uvx steamworks-mcp setup
   

2. Type the folder that holds your games (Enter keeps the one shown). The keys it offers are optional: Enter skips them.

3. When it asks Codex?, answer y. If it says "not found", Codex has not run on this computer yet; y still works.

4. Restart Codex (the app, or a new CLI session) and ask: "Where are my games on Steam?"

The server goes into ~/.codex/config.toml, which the Codex app, the CLI and the IDE extension all read, so one setup covers all three. Codex starts the server on your computer: no tunnel and no access code. Tools that only read run without asking; Codex asks before every tool that changes something.

Check the installation

uvx steamworks-mcp doctor

Other apps (Cursor and anything else that supports MCP) and connecting by hand: docs/ADVANCED.md.

Your first conversation

Just talk to your assistant. Some things you can say:

Every answer starts with one plain sentence and, where it helps, a table. The assistant shows you each draft; only you approve.

Words you will see

What it can do

Start and see where you are

Store page, images and translations

Market (live data from Steam's public store)

Your game's code

Steamworks

How each area gets done in Steamworks (Web API, BROWSER mode, a file to upload, or a manual step) and how well it is verified: docs/CAPABILITIES.md.

Skills

36 recipes for your assistant. In Claude Code with the plugin they show as /steamworks:<name>.

Each skill uses the server's tools where one fits and links Valve's documentation. A test checks that every tool, steamworks.yaml field and skill a skill names really exists.

Code rules

check_code reads your game's own code (plugins and the SDK wrappers are skipped) and reports each problem with the file, the line, why it matters and the fix. A key or password it finds is never shown.

The full list with Valve's sources: src/steamworks_mcp/data/code_rules.yaml.

Safety: what it never does

• Publish, prepare to publish, revert, submit for review or release anything. You do that in Steamworks. (Store tags go live as soon as they are saved; the tool saves them only with your separate goes_live_now=true.)
• Write anything to Steamworks without a dry run first and your OK, or write values you have not approved.
• Edit packages (they change what customers own at once) or set a build live on the default branch.
• Delete anything in Steam unless you explicitly ask for it.
• Type, store or ask for passwords or Steam Guard codes, or read cookie values.
• Write into your game project. The scanners only read; generated code goes to .steam-mcp/exports/ for you to copy.
• Generate artwork. Images are cut from your own art; missing art is reported.
• Send your texts to a translation or AI service. Your own assistant writes and translates; the server checks.

Every write saves a snapshot of what Steam had, reads Steam again afterwards and is logged. Details: Writing to Steam and BROWSER mode.

Known limits

• Scans read Unity, Godot and Unreal projects. Other engines answer the questions instead; everything after the scan works the same.
• Still manual in Steamworks: pricing, the content survey and ratings, the Controller and Accessibility surveys, the release date, stat definitions, screenshots and trailers, creating depots, packages and DLC, and setting a build live on the default branch. The checklist from export_package tells you exactly where.
• Not verified live: writing store tags, setting a beta branch live, the app and shortcut icons, and the Auto-Cloud overrides suggested for macOS and Linux saves.
• Images: the tool fills empty image slots only; it never replaces or removes an image in Steamworks.
• The BROWSER mode follows the Steamworks site as it was on 2026-10-03. When a page changes, the tool stops before writing.
• Market numbers come from Steam's public store; Steam publishes no sales, so sales estimates are rules of thumb.

More

• docs/ADVANCED.md: configuration, keys and accounts, connecting each app by hand, remote clients, how writes work, the BROWSER mode, files, development.
• docs/CAPABILITIES.md: what is automated in each Steamworks area, and how well it is verified.
• docs/SCHEMA.md: every field of steamworks.yaml.
• examples/example-game/steamworks.yaml: a complete example.

License

MIT