AGS API MCP Server

Last updated on October 6, 2026

Overview

The AGS API MCP Server connects your AI assistant to AccelByte Gaming Services (AGS). Once connected, you can ask your assistant questions like "Which endpoint grants an item to a player?" or "List the stat codes in my namespace," and it answers from your real AGS environment instead of guessing.

AccelByte hosts the server for you. You don't install or run anything. You give your assistant one URL, log in with your AGS account, and you're done.

This page builds up in three stages. Stop at whichever stage gives you what you need:

Connect: add the server to your assistant and log in.
Try it: run a few prompts, from read-only lookups up to real changes.
Go further: charts, what your requests reveal, self-hosting, and fixes for common problems.

What MCP is

The Model Context Protocol (MCP) is an open standard that lets AI assistants use outside tools. An MCP server offers a set of tools. Your assistant decides when to call them. Here, the tools let your assistant search the AGS API catalog, read the details of an endpoint, and send requests to AGS on your behalf. Most current assistants support MCP, including Claude, Cursor, Visual Studio Code with GitHub Copilot, Gemini CLI, and Antigravity.

Before you begin

You need:

An AGS account you can log in to the Admin Portal with.
An AI assistant that supports MCP. The setup steps below cover Claude Desktop, Claude Code, Cursor, Visual Studio Code, Antigravity, and Gemini CLI.

That's all for now. A few assistants need one extra piece (Node.js), and the steps tell you if yours does.

note

Your assistant can only do what your AGS account can do. The server acts as you, with your roles and permissions. If you can't do something in the Admin Portal, your assistant can't do it either.

Stage 1: Connect

You can connect in one of two ways:

Let your assistant do it. Paste one prompt and your assistant handles the setup.
Do it yourself. Follow Steps 1 to 3 below.

Let your assistant do it

This works in assistants that can edit files on your computer: Claude Code, Cursor, VS Code, Antigravity, and Gemini CLI. Paste this prompt into your assistant:

Install the AGS API MCP server for me. Fetch and follow the instructions at
https://raw.githubusercontent.com/AccelByte/ags-api-mcp-server/refs/heads/master/INSTALL.md

Your assistant asks for your AGS edition and namespace, builds your URL, writes the config, and checks that the connection works. When it's done, skip to Stage 2: Try it.

tip

Using Claude Desktop? The manual path is faster for you. Follow Steps 1 to 3 below and choose the Claude Desktop tab in Step 2.

Step 1: Find your server URL

Your URL depends on which AGS edition you use. If you're not sure, ask your AccelByte administrator.

Public Cloud
Private Cloud

Your URL uses your studio namespace twice, once at the start and once at the end:

https://<YourStudioNamespace>.prod.gamingservices.accelbyte.io/mcp/<YourStudioNamespace>

For example, if your studio namespace is mystudio, your URL is:

https://mystudio.prod.gamingservices.accelbyte.io/mcp/mystudio

To find your studio namespace, log in to the Admin Portal and check the address bar. You can also find it in the welcome email you received when you registered.

tip

Use your studio namespace only. Don't add a game namespace to the URL. Studio admins and game admins connect to the same URL, and your roles decide which games you can reach.

info

Older guides and support threads call Public Cloud "Shared Cloud." They are the same thing.

Your URL is your environment's domain followed by /mcp:

https://<YourDomain>/mcp

For example, if your environment's domain is games.example.com, your URL is https://games.example.com/mcp.

note

The MCP endpoint might not be turned on for your environment yet. If your assistant can't connect, contact AccelByte support to enable it.

Copy your URL. You'll paste it in the next step wherever you see <YourMcpUrl>.

Step 2: Add the server to your assistant

Pick your assistant. Each tab shows the shortest path that works.

Open Settings > Connectors, then select Add custom connector.
In Name, enter ags-api.
In Remote MCP server URL, paste your URL.
Select Add.

Claude Desktop handles the login for you. You don't edit any files.

Don't see Add custom connector?

Some Team and Enterprise plans turn off custom connectors. In that case, add the server through the config file instead. This path needs Node.js 18 or later. Run npx --version in a terminal to check that you have it.

Open claude_desktop_config.json:
- macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
- Windows: %APPDATA%\Claude\claude_desktop_config.json

Add the ags-api entry inside mcpServers:

{
  "mcpServers": {
    "ags-api": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "<YourMcpUrl>"]
    }
  }
}

Save the file, then fully quit and reopen Claude Desktop.

Run this in your terminal:

claude mcp add --transport http ags-api <YourMcpUrl>

Open .cursor/mcp.json in your project. Create it if it doesn't exist.

Add the ags-api entry inside mcpServers:

{
  "mcpServers": {
    "ags-api": {
      "type": "http",
      "url": "<YourMcpUrl>"
    }
  }
}

Save the file and reload the Cursor window.

Open .vscode/mcp.json in your project. Create it if it doesn't exist.

Add the ags-api entry inside servers:

{
  "servers": {
    "ags-api": {
      "type": "http",
      "url": "<YourMcpUrl>"
    }
  }
}

Save the file and reload the VS Code window.

note

VS Code uses servers as the top-level key. Other assistants use mcpServers. If you copy a config from another tool, rename the key.

Open mcp_config.json in your project root. Create it if it doesn't exist.

Add the ags-api entry inside mcpServers:

{
  "mcpServers": {
    "ags-api": {
      "type": "http",
      "url": "<YourMcpUrl>"
    }
  }
}

Save the file and reload the Antigravity window.

Run this in your terminal:

gemini mcp add --transport http ags-api <YourMcpUrl>

tip

If your config file already lists other MCP servers, add ags-api next to them. Don't replace the whole file, or you'll lose your other servers.

Step 3: Log in and check the connection

Ask your assistant: "What's my current user information?"
A browser window opens to the AccelByte login page. Log in with your AGS account and approve the consent screen.
Back in your assistant, you see your user details, namespace, roles, and when your session expires.

If you see your details, you're connected. Your assistant reuses this login for future requests, so you won't need to log in every time.

If the browser doesn't open or the login fails, go to Troubleshooting.

Stage 2: Try it

Your assistant now has four tools. You don't call them by name. Ask in plain language and your assistant picks the right one.

Try these prompts in order. Each one goes a step further than the last.

Step	Try asking	What happens	Changes anything?
1	"What's my current user information?"	Shows who you're logged in as and what you can access.	No
2	"Find the APIs for managing player inventories."	Searches the AGS API catalog by description, service, or method.	No
3	"What parameters does the endpoint for granting an item need?"	Reads the full details of one endpoint: parameters, request body, and required permissions.	No
4	"List the stat codes in my game namespace."	Sends a read request to your AGS environment and shows the result.	No
5	"Create a stat code called `kills` in my game namespace."	Sends a write request. Your assistant asks for your approval first.	Yes

warning

Requests in step 5 change real data in your environment. Your assistant always asks before it sends a create, update, or delete request. Read each request before you approve it, especially in a production namespace.

For a change that spans several steps, use the AGS CLI instead. It shows the whole plan before anything runs. See Which tool to use.

tip

Be specific about the namespace. If your account can reach several games, name the one you mean, for example, "in the mystudio-mygame namespace."

Stage 3: Go further

Turn query results into charts

The server can draw charts, tables, and single-number metrics right inside assistants that support interactive MCP content. Combined with AGS analytics data, you can ask questions like "Show daily active players for the last 30 days as a line chart."

To set this up, including the permissions and playbooks you need, see Getting started with AGS AI Analytics.

Know what your requests reveal

The server has no separate telemetry to opt out of. Each request it sends to AGS is an ordinary API call made with your login. The server adds only a User-Agent header that names itself and your assistant, for example:

ags-api-mcp-server/2026.1.0 (MCP-Client claude-code/2.1.0)

This header lets your administrators tell requests made through your assistant apart from requests made in the Admin Portal. Analytics queries also carry the version of the playbook and game context they used, and nothing else.

note

What AGS returns goes back to your AI assistant, so your assistant's provider processes it like anything else you share in a chat. Keep this in mind before you ask for player data. Your organization's AI policy applies here as it does to any other chat.

Run the server yourself

You don't need to host the server, because AccelByte runs it for you. If your team wants to run its own copy anyway, follow the self-hosting guide in the AGS API MCP Server repository.

Troubleshooting

Find the symptom that matches yours.

The login fails with an OAuth, DCR, or "registration failed" error

Your assistant doesn't support automatic client registration yet. Switch to the bridge config, which handles the login for your assistant. It needs Node.js 18 or later. Run npx --version to check that you have it.

Cursor, VS Code, Antigravity: replace the ags-api entry in your config file with:

"ags-api": {
  "command": "npx",
  "args": ["-y", "mcp-remote", "<YourMcpUrl>"]
}

Claude Code: remove the server with claude mcp remove ags-api, then run claude mcp add ags-api -- npx -y mcp-remote <YourMcpUrl>.
Gemini CLI: remove the server with gemini mcp remove ags-api, then run gemini mcp add ags-api -- npx -y mcp-remote <YourMcpUrl>.

Restart your assistant after the change.

The browser doesn't open when you ask your first question

Check that you saved the config file.
Fully quit and reopen your assistant. Reloading the window isn't always enough.
Ask "What's my current user information?" again.

The login worked before, but now fails with "Invalid Request" or "client ID not found"

Your assistant saved a registration that AGS has since removed, for example, after an unused client was cleaned up. Clear the saved login for the ags-api server only, then log in again:

Claude Code: run /mcp, select ags-api, and choose Clear authentication. Then reconnect.
Codex: run codex mcp logout ags-api, then codex mcp login ags-api.
Bridge config (mcp-remote): follow the steps in the next item.

You don't need to remove and re-add the server.

The browser opens for login every time

The bridge config (mcp-remote) stores logins in the ~/.mcp-auth/ folder. If that folder is stale or can't be written to, you get logged out on every launch.

Check that your user account can write to ~/.mcp-auth/.
If it can, rename the folder as a backup and log in again:
```
mv ~/.mcp-auth ~/.mcp-auth.bak
```
If that doesn't help, restore the backup with mv ~/.mcp-auth.bak ~/.mcp-auth.

note

Renaming this folder logs you out of every MCP server that uses the bridge config, not just this one.

You're logged in, but requests fail with a 403 error

Your AGS account doesn't have permission for that request. This isn't a setup problem. Ask your AGS administrator to grant you the role you need.

Still stuck? Check the AGS API MCP Server repository for the latest known issues, or contact AccelByte support.

Overview​

What MCP is​

Before you begin​

Stage 1: Connect​

Let your assistant do it​

Step 1: Find your server URL​

Step 2: Add the server to your assistant​

Step 3: Log in and check the connection​

Stage 2: Try it​

Stage 3: Go further​

Turn query results into charts​

Know what your requests reveal​

Run the server yourself​

Troubleshooting​

Overview

What MCP is

Before you begin

Stage 1: Connect

Let your assistant do it

Step 1: Find your server URL

Step 2: Add the server to your assistant

Step 3: Log in and check the connection

Stage 2: Try it

Stage 3: Go further

Turn query results into charts

Know what your requests reveal

Run the server yourself

Troubleshooting