Catalog›Plugins›accelbyte-ai-pluginsAAaccelbyte-ai-pluginsPublic AI coding agents, skills, and MCP servers for AccelByte. Open site Project Open in the app JSON README (API)READMELoading documentation…# accelbyte-ai-plugins  Public AI coding agents, skills, and MCP servers for AccelByte. ## How it works Once installed, skills activate automatically when your AI assistant recognizes a relevant AccelByte question — no special commands needed. You can also invoke any skill directly using its slash command. For connecting your assistant to a live AccelByte environment, each skill can configure MCP servers on a per-project basis — your assistant will walk you through it when you're ready. ## Usage telemetry When the `ags`, `ags-extend` or `teammate` skill activates, it sends one anonymous event to AccelByte's PostHog at `e.accelbyte.io`. The event is named `skill_activated`. Its properties are `skill` and `plugin_version`, plus two PostHog flags that turn off person profiles and GeoIP lookup. The request's User-Agent is `AccelByte-AI-Plugins/<plugin version> (skill=<skill name>)`. The request carries no prompts, code, file names, repository or project details. We send no user, device, or installation identifier. Every installation sends the same `distinct_id`, and PostHog builds no profile or location from it. PostHog receives your IP address with the request, as any server does, and discards it instead of storing it with the event. The call is best-effort. It times out after 2 seconds, is never retried, and a failure never affects the skill. Other assistants may require approval for shell commands or restrict network access. Telemetry may not be sent under those policies. The skill must skip telemetry if it would require additional permissions and must not retry it outside the sandbox. To turn it off, set `ACCELBYTE_AI_PLUGIN_TELEMETRY=0` (`false` and `off` also work) in the environment your assistant runs in. Setting `DO_NOT_TRACK` to any non-empty value, which also turns off the AGS CLI's telemetry, turns it off too. The script that sends it is `skills/accelbyte/telemetry/telemetry.sh`. The Kiro skills send nothing, because Kiro would ask you to approve the call every time. ## Skills ### accelbyte The `accelbyte` skill is the main router for this plugin. It recognizes AccelByte-related prompts, sends AGS work to `/ags`, sends Extend work to `/ags-extend`, and keeps answers grounded in bundled references. Use the AGS and Extend sections below for the actual start commands and workflow details. ### ags AccelByte Gaming Services (AGS) is the managed backend that handles player accounts, sessions, matchmaking, leaderboards, achievements, the item store, wallet, and more — so you can focus on building your game. > [!NOTE] > - **/ags init** - start here for a clean-slate setup. It scans the current project, detects the engine or app type, and guides the setup for AGS configuration, SDKs, the AGS CLI, MCP servers, namespace access, and IAM client settings > - **/ags <prompt>** - use this for every question, goal, or symptom after that, such as `/ags help me add login and matchmaking` or `/ags debug why tickets are timing out`. #### What it covers The full AGS platform: player authentication (IAM), real-time lobby and presence, sessions and matchmaking, leaderboards, achievements, item store, wallet, entitlements, and namespace management. #### Examples ``` /ags what AGS modules do I need for a co-op shooter? ``` ``` /ags init ``` ``` /ags help me add login and matchmaking to this Unreal project ``` ``` /ags debug why matchmaking tickets are timing out ``` ``` /ags check my namespace setup before I change anything ``` ``` /ags do I need AMS or Extend for this? ``` ``` /ags add AGS login to this game ``` ``` /ags set up matchmaking for a 4-player co-op mode ``` ``` /ags help me add achievements and leaderboards ``` #### What `/ags` can help with - **Understand AGS** - choose the right modules for auth, lobby, matchmaking, sessions, store, achievements, live ops, and more. - **Set up a project** - bootstrap namespace access, IAM client config, SDK install, CLI setup, and optional MCP configuration. - **Integrate game features** - wire AGS into Unreal, Unity, Godot, Roblox, web, or a custom engine. - **Debug issues** - trace auth failures, lobby disconnects, matchmaking timeouts, store problems, and namespace/config mistakes. - **Review safely** - inspect an existing namespace or diagnose a symptom before changing anything. - **Plan multiplayer depth** - design or troubleshoot matchmaking rules, ticket flow, region routing, sessions, and dedicated-server use. - **Decide boundaries** - know when to stay in AGS, use AMS, add Extend, consider ADT, or involve AccelByte support. ### ags-extend AGS Extend lets you run custom server-side game logic — matchmaking filters, leaderboard tiebreakers, anti-cheat hooks, custom statistics, and more — as lightweight services deployed on AccelByte's infrastructure, without managing your own servers. #### What it covers Picking the right Extend pattern, scaffolding a new service, defining its API, running it locally, deploying it to AGS, and debugging it in production. #### Examples ``` /ags-extend What Extend pattern fits a custom matchmaking filter? ``` ``` /ags-extend Scaffold a new Override service for leaderboard ranking ``` ``` /ags-extend My service won't start after deploy — help me debug it ``` #### What you can do - **Plan your service** — `/ags-extend wizard` helps you pick the right pattern (Override, Custom Function, Event Handler) for your use case. - **Scaffold a project** — `/ags-extend init` sets up a new service from an official template, with dependencies and tooling ready. - **Define the interface** — `/ags-extend proto` helps you write the protobuf definition for your service's API. - **Run and test locally** — `/ags-extend test` runs your service against a local AGS environment before you deploy. - **Deploy to AGS** — `/ags-extend deploy` packages and pushes your service to your AGS namespace. - **Check logs and health** — `/ags-extend observe` and `/ags-extend debug` trace what's happening once it's live. - **Set up CI** — `/ags-extend ci` wires automated builds and deploys into your pipeline. #### Intended workflow 1. **Not sure where to start?** `/ags-extend ask` explains Extend patterns and when to use each one. 2. **Planning a multi-app project?** `/ags-extend design` helps you shape the architecture before writing a line of code (optional for single-service work). 3. **Starting a service** — `/ags-extend init` does the full setup: scaffolds a template, installs dependencies and the CLI, and optionally sets up MCP servers. Or step through it: `/ags-extend wizard` → `/ags-extend install-dep` → `/ags-extend install-cli`. 4. **Iterating** — `/ags-extend proto` after SDK or contract changes, `/ags-extend debug` to run locally, `/ags-extend test` for unit, integration, and contract tests. 5. **Ready to ship** — `/ags-extend deploy` builds, pushes, and deploys to AGS. `/ags-extend ci` wires it into GitHub Actions or GitLab CI. 6. **In production** — `/ags-extend observe` for logs and health, `/ags-extend doctor` if something's off, `/ags-extend upgrade` for SDK or proto version bumps. ### teammate An AI colleague for your AccelByte integration. Point it at your game repo and it checks how your AGS integration is wired, then hands you a report where every finding is backed by a citation or left out. #### What it covers Your repo's AccelByte SDK usage: incomplete integrations, deprecated APIs, unsafe token handling, and calls with no path for failing — surfaced as a report you can act on, held to a grounded-or-suppressed rule so no claim ships without a citation. It also answers two questions that are not about the code as it stands: what an engine SDK upgrade would break, and whether one AMS fleet or Extend app is sized right. For an AMS fleet it goes further than sizing — one named fleet's configuration, its image and artifact hygiene, and what the last day of its server history says about crashes — and for an Extend app it goes further too: whether it is running, whether its last deployment worked, whether the image it serves carries critical findings, and whether the configuration in the Portal is the configuration in force. When one of those stops — a dedicated server that died, a deployment that failed — it builds the timeline and names the cause. And it keeps what the code never records — hand it a design, a plan or a postmortem and it files the text in your studio's own memory, so a later answer can reach what the team decided. #### Examples ``` /teammate health-check ``` ``` /teammate check my Extend app CPU and memory usage and advise the optimal settings ``` ``` /teammate check my prod-eu fleet ``` ``` /teammate is my matchmaking-override Extend app healthy ``` ``` /teammate why did server ds-01j2 die at 14:20 ``` ``` /teammate remember this technical design ``` ``` /teammate what can you check? ``` #### What you can do - **Health check** — `/teammate health-check` — scan an AccelByte-integrated repo for integration gaps, deprecations, and auth-token-safety issues, and get a cited report. - **Upgrade check** — `/teammate upgrade-check` — for a version bump of your engine SDK, see which of your own call sites break, each at `file:line`. It reads only. - **Sizing check** — `/teammate sizing-check` — for one named AMS fleet or Extend app, see what it is set to and what it should be, with each number labelled by what it rests on. - **Fleet check** — `/teammate fleet-check` — for one named AMS fleet, see what is misconfigured, what its image and artifact hygiene misses, what its last day of server history says about crashes, and what min, max and buffer each region should carry. - **Extend app check** — `/teammate extend-app-check` — for one named Extend app, see whether it is running, what its last deployments did, whether its active image carries critical findings, which variables and secrets are not yet in force, and what its CPU, memory and replicas should be. It never reads a secret's value back to you. - **Why did it die** — `/teammate why-did-it-die` — for one dedicated server or one Extend app deployment that stopped, see what happened in the order it happened, which cause fits, and where the log or core dump is. It hands the artifact over; it does not read it for you. - **Remember** — `/teammate remember` — hand over a technical design, a milestone plan, meeting notes or a postmortem and it files the text, unedited, in your studio's memory. Needs the memory server; without one it stores nothing and says so. - **Ask** — `/teammate` — ask what the teammate can do, or what past scans already found, and get routed to the right check. #### Intended workflow 1. Install this skill (optionally add its memory MCP for report reuse and for `remember`). 2. Open your AccelByte-integrated game repo. 3. Run `/teammate health-check` to scan it. 4. Review the report — every finding is cited or left out. 5. Act on the findings, then re-run to confirm. 6. Hand it the designs and plans behind the code with `/teammate remember`, so a later answer reaches them too. ## MCP Servers ### AGS API MCP Server Connects your AI assistant directly to your live AGS environment. Instead of guessing, it can read your actual namespace config, check API responses, and give advice grounded in your real setup. Useful when debugging a live integration or validating how your game's backend is actually configured. ### AGS Extend SDK MCP Server Gives your AI assistant direct access to AccelByte Extend SDK types, functions, and models while you code. Supports Go, Java, Python, and C#. Instead of hallucinating API shapes, your assistant can look them up — making AI-generated Extend code significantly more accurate. ### Teammate Memory MCP Server Remembers your studio's scans and the documents you hand it, so a report can be reused, a design or a postmortem stays readable, and a colleague's activity can surface while you work. Optional, though `/teammate remember` needs it: without it the teammate still scans and reports, it just can't keep a document, reuse a prior report, or see colleague activity. ### Teammate Studio Wiki MCP Server Serves your studio's own scans and documents as readable pages, so a run can orient itself on what your team already found and on what it decided. Optional, and it starts empty: it serves what has been rewritten so far. Nothing here ever grounds a claim about AccelByte. ### AccelByte Unity MCP Adds Unity-specific AccelByte tooling, starting with deterministic AGS uGUI prefab generation. Your assistant can discover project style, inspect AGS kit/spec drift, validate shared AGS recipes, resolve Unity prefab specs, and generate typed TMP/uGUI screens through the live editor bridge or Unity batch mode. ### AccelByte Unreal SDK MCP Server Indexes the AccelByte Unreal SDK - classes, methods, code snippets, and ready-made Slate UI panels for Login, Achievements, and Matchmaking. Your assistant can search real SDK symbols and examples instead of guessing, making Unreal Engine integration with AccelByte significantly more reliable. ## Installation ### Claude Code **CLI:** ``` /plugin marketplace add AccelByte/ai-plugins /plugin install accelbyte-ai-plugins@accelbyte ``` **From within Claude Code:** 1. Run `/plugin` and navigate to the **Marketplaces** tab. 2. Select **Add Marketplace**, enter `AccelByte/ai-plugins`, and confirm. 3. Install the **accelbyte-ai-plugins** plugin. 4. Run `/reload-plugins` to activate the plugin. ### Claude Desktop **Automatic (Cowork) — paste this prompt into a Cowork chat:** ``` Download `https://github.com/AccelByte/ai-plugins/archive/refs/heads/main.zip`, unzip it, remove the `mcpServers` and `userConfig` fields from .claude-plugin/plugin.json (Cowork's plugin validator does not support these fields yet), repack it as a .zip with all contents at the archive root — if the zip contains a single top-level directory (as GitHub typically adds), strip that wrapper so files like `.claude-plugin/` appear directly at the root — rename it to `accelbyte-ai-plugins.plugin`, and use the `present_files` tool to present it to me. ``` **Manual (Chat and Cowork):** 1. Download the repo archive: [AccelByte/ai-plugins/archive/refs/heads/main.zip](https://github.com/AccelByte/ai-plugins/archive/refs/heads/main.zip) 2. In Claude Desktop, open **Customize → Upload plugin** and select the downloaded file. 3. Confirm the install. ### Codex **CLI:** ``` codex plugin marketplace add AccelByte/ai-plugins ``` **From within Codex:** 1. Run `/plugins` and navigate to the **Add Marketplace** tab. 2. Enter `AccelByte/ai-plugins` and confirm. 3. Install the **accelbyte-ai-plugins** plugin. 4. Restart Codex to activate the plugin. ### Any agent (Agent Skills) ``` npx skills add AccelByte/ai-plugins ``` ### Cursor, Kiro, OpenCode Paste this into your AI assistant: ``` Fetch and follow instructions from https://raw.githubusercontent.com/AccelByte/ai-plugins/refs/heads/main/INSTALL.md ``` --- Built with AccelByte External Marketplace compiler `v0.9.2`. DetailsTypePluginsPublisheraccelbyteSourcemarketplaceCategoryToolsStars4Copies (forks)3Proposed changes1Last updateNov 10, 20265 days agoProject statusActiveLanguageTypeScriptAddedAug 30, 20263 months agoUpdatedAug 30, 20263 months agoSource idaccelbyte/ai-plugins/accelbyte-ai-pluginsMorePluginsAll from accelbyteA–Z index
