Edgegap のリモート MCP サーバーとは何ですか?
Edgegap のリモート MCP サーバーは、https://mcp.edgegap.dev/mcp にホストされた Model Context Protocol エンドポイントです。AI アシスタントはローカルに何もインストール・実行することなく接続できます。
Edgegap MCP Server認証済み注目
@Mathieu Duperre
基本情報
Edgegap MCP Server について
edgegap-mcp は、390 以上のスタジオのライブマルチプレイヤータイトルを支えるオーケストレーションプラットフォーム Edgegap に、コーディングエージェントを接続します。作業の途中でダッシュボードや API リファレンスに移る代わりに、エージェントがアプリケーションを作成し、コンテナイメージを登録し、プレイヤーの近くにデプロイし、起動を待って、接続アドレスを返します。 コンテナで動く専用サーバーなら何でも使えます：Unity、Unreal Engine、Godot のプロジェクトに加え、独自の Go、Rust、Node のバックエンドも。ビルドがヘッドレスサーバーと Dockerfile を生成できれば、デプロイできます。 10 個のツールは、API 仕様から自動生成したものではなく厳選したものです。基本の流れをカバーします：アプリケーションの一覧と作成、CPU、メモリ、ポートを指定したバージョンの登録、デプロイ、バックオフ付きでの起動待ちポーリング、クラッシュ後のコンテナログ取得、きれいな停止。エラーはエージェントが自ら修正できるように書かれています。 npm パッケージ @edgegap/mcp としても提供されており、ローカルで動作し、無人のエージェント向けに読み取り専用モード、アプリケーションの許可リスト、デプロイ時間の上限を追加します。 Edgegap は任意の Docker コンテナイメージを、どこでも、1 秒以内にオンデマンドで実行できます。
接続情報
https://mcp.edgegap.dev/mcpセットアップ
claude mcp add edgegap --transport http https://mcp.edgegap.dev/mcp --header "Authorization: token YOUR_API_TOKEN"Edgegap MCP Server のツール
32Report which Edgegap API this server talks to, its version, whether it is the local or hosted server, and the guardrails in force: read-only mode, application allowlist, max_duration ceiling, the agent deployment cap, the tag forced onto every deployment, and which tools support dry_run. Call it first when unsure what this session is allowed to do, or before anything billable. Makes no API call and never shows the token.
List the applications in the Edgegap organization. Start here before creating or deploying anything, so you reuse an existing application instead of making a duplicate. An "application" groups versions of one game server.
Create a new application to hold game server versions. Only call this after edgegap_list_apps confirms no suitable application exists. Creating an application does not deploy anything — follow with edgegap_create_app_version.
Set whether an application accepts new deployments. Turning it off stops new deployments of every version under it; running deployments keep running. Use it to pause an application without deleting anything.
List the versions under an application, with their container image, resources, ports (including whether TLS Upgrade is on, which browser/WebGL clients need for wss://) and whether a registry login is set for pulling the image. Use this to find the version name to deploy, to check a version is ready before deploying, or to copy settings from a working version when creating a new one.
Read one version in full: image, resources, ports, max_duration, whether image caching (force_cache) and verify_image are on, restart policy, and the names of its env vars (values are never shown). Use it before edgegap_update_app_version or edgegap_deploy to check what a version will do.
Register a container image as a deployable version of an application. The image must already be pushed to a registry that Edgegap can pull from (see edgegap_get_registry_credentials and edgegap_list_registry_tags). Resource units: 1024 cpu units = 1 vCPU; memory_mb must be at least 256 and at most double the cpu units. Set verify_image true on the first version so a bad image fails here rather than at deploy time. Refused, with nothing sent, for the "latest" or an empty docker tag (use a build ID so deployments are reproducible) and for a docker_image that also holds a registry host or tag. Env vars named like secrets (TOKEN, SECRET, PASSWORD, API_KEY…) are hidden in the dashboard.
Change settings on an existing version instead of creating a new one: point it at a new image tag, turn it on or off, change max_duration, set the registry login, turn on image caching (force_cache: Edgegap pre-loads the image so deployments start faster), or set the shutdown grace period. Only the fields you pass change. Deployments already running keep their old settings; new deployments get the new ones. Same refusals as edgegap_create_app_version: "latest" or empty tags, a docker_image holding a host or tag, and the API token in the registry login. Resources, ports and env vars cannot be changed here; create a new version for those.
List the cities Edgegap can deploy to. Pass application and version to see only the locations with capacity for that version right now. Use it to pick geo_coordinates for edgegap_deploy when you have no player IPs, or to tell a developer where their players can be served from. For the ping beacons game clients use to measure latency, see edgegap_list_ping_beacons.
List Edgegap's ping beacons: one host per region that a game client can ping (UDP or TCP) to measure its latency before matchmaking. The matchmaker's latency rules use these measurements. Read-only; the game client does the pinging, not this tool.
Start one authoritative dedicated game server from an application version, placed near the players you specify. This is the recommended way to host a multiplayer match: the server owns the game state, so no player hosts it. Returns immediately with a request_id; the server is still starting and has no connection details yet. Follow this call with edgegap_wait_for_deployment to get the address players connect to. Always stop deployments you started for testing. Every deployment is tagged "edgegap-mcp" in addition to any tags you pass, so edgegap_list_deployments with only_agent_deployments: true finds them all. Runs until stopped or until the version's max_duration, which the result reports. Refused, with nothing sent, if the API token appears in an env var or tag, if the version uses the "latest" docker tag, or when too many agent deployments are already running (EDGEGAP_MAX_AGENT_DEPLOYMENTS, default 5): stop finished ones first. Env vars named like secrets are hidden in the dashboard.
Read the current status of one deployment, including connection address and ports once it is ready. For a deployment you just created, prefer edgegap_wait_for_deployment — it polls for you instead of making you call this in a loop.
Poll a deployment until it is ready, errors, or the timeout expires, then return the connection details. This is the tool to call right after edgegap_deploy. Do not build your own polling loop — this handles backoff and reports the container error detail if the server fails to start.
List active deployments, optionally filtered and sorted. Use this to find deployments left running from earlier sessions before starting new ones — orphaned servers cost money. Filter by status, application, version, tags, request_id, created_at, fleet_name or host_name, e.g. [{field:"application",operator:"eq",value:"my-game"},{field:"status",operator:"eq",value:"ready"}]. Operators: eq and neq on every field except created_at (eq, gte, lte); in and nin with an array value on request_id, tags, application, version, fleet_name, host_name; ilike with % wildcards on fleet_name and host_name. At most one filter per field. Sort by created_at or available_session_sockets. To find every deployment an agent started through this server, pass only_agent_deployments: true (the same as filtering tags in ["edgegap-mcp", "mcp"]).
Gracefully stop one deployment by request_id, sending SIGTERM to the container. Stop every deployment you started for testing before ending your task. This tool stops exactly one deployment, named by its exact request_id: empty IDs, wildcards and lists are refused, and bulk stop is deliberately not exposed.
Retrieve stdout/stderr and crash output for a deployment. Call this whenever a deployment errors or a server exits unexpectedly — the crash exit code usually identifies the problem faster than redeploying does. Logs for stopped deployments are only retained if Endpoint Storage was configured on the version beforehand.
Summarise a deployment's resource use over time: CPU, CPU throttling, memory and network, each as latest, average and peak, with the API's own labels. Use it to size a version (cpu_units, memory_mb) from a real match, or to check whether a crash was memory or CPU starvation before redeploying. Defaults to the whole life of the deployment.
Write a Dockerfile for this project's headless game server build, ready for Edgegap: linux/amd64 base, the build folder copied in, the binary made executable, the right headless flags (Unity -batchmode -nographics, Godot --headless), a non-root user where the engine needs it (Unreal refuses root), CRLF fixes for start scripts, and EXPOSE lines matching the ports. Call this when the project has no Dockerfile, before building anything. Look at the build output first so you can pass the real build folder, binary name and port; anything you omit is assumed and listed under assumptions, which you must confirm with the developer or the project files. Write the result to Dockerfile, then build. The output is checked against edgegap_validate_server_config before it is returned. Makes no API calls.
Check a game server Dockerfile and the ports/resources you intend to register against what Edgegap requires, BEFORE building and pushing. Catches the failures that otherwise only show up after a build, push, version and deploy: ARM or Windows images (Edgegap runs linux/amd64), Unreal running as root, missing Unity -batchmode -nographics, a server bound to localhost, EXPOSE ports that do not match the version ports, a protocol that does not match the netcode transport, the "latest" tag, and bad CPU/memory ratios. Pass the Dockerfile text (read it from disk first). If there is no Dockerfile yet, call edgegap_generate_dockerfile instead. Makes no API calls.
Get the push login for this organization's private Edgegap container registry (registry.edgegap.com) and the exact docker login, build and push commands for your image. Use this when the server image is not in a registry yet. Edgegap only hands these credentials to the quick-start token the Unity/Unreal plugins use; with a regular API token this tool cannot fetch them and instead returns the steps to copy them from the dashboard's Container Registry page — ask the developer for them then, do not retry. When it does return a token, it is registry-scoped (push/pull images in this project), not the org API token: pass it to docker login via --password-stdin, never as a command-line argument, and do not write it into files. Call edgegap_validate_server_config before building.
List the tags pushed for one image in this organization's Edgegap container registry, with push time and size. Call it after docker push to confirm the tag landed before edgegap_create_app_version, which otherwise fails later with an image-pull error.
Delete one tag, and the image it points to, from this organization's Edgegap container registry, to clean up old builds. Takes exactly one image name and one exact tag; wildcards, lists and "latest" are refused, and there is no bulk delete. A version that still uses the tag can no longer start deployments, so check edgegap_list_app_versions first. Cannot be undone.
A relay is NOT a game server and is not the recommended default — for a multiplayer match, deploy an authoritative dedicated server with edgegap_deploy instead. A relay only forwards traffic between players while one player's game acts as host and owns the game state: the host can cheat and has a latency advantage, the host's PC and connection cap the match, and the match ends if the host quits. Only use this when the developer has chosen peer-to-peer/host-client, or the netcode is already a listen server and a dedicated server is not an option; if that has not been established, ask first. Never pick it only because it needs no server image. Creates an Edgegap relay session for every player's public IP, host first, waits until it is ready, and returns the relay address plus per-player authorization tokens for the relay transport. Relay sessions are billed while open: delete test sessions with edgegap_delete_relay_session when finished.
Read one relay session: whether it is ready, the relay address and ports, and each authorized player with their authorization token. edgegap_create_relay_session already waits for readiness; use this to re-read a session or one created with wait_until_ready false.
List this organization's active relay sessions: id, status, readiness and player count. Relay sessions bill while open, so use this to find ones left over from earlier sessions and delete them with edgegap_delete_relay_session. Authorization tokens are not listed; read one session with edgegap_get_relay_session.
Authorize one more player (by public IP) on an existing relay session, for a player joining after the session was created. Returns that player's authorization token.
Revoke one player's access to a relay session, by that player's authorization token (from edgegap_get_relay_session or edgegap_authorize_relay_user). The player loses the relay connection; the session and the other players stay.
Close one relay session. Connected players lose their relay connection. Delete every session you created for testing before ending your task. Takes exactly one session_id; empty IDs, wildcards and lists are refused.
Generate a ready-to-upload Edgegap matchmaker JSON configuration with one profile: team count and size, optional latency rule, and optional expansions that relax the rules the longer a player waits. Checks the referenced application version exists and has ports. Edgegap has no API for creating a matchmaker, so this tool does NOT create one: save the returned config to a file (e.g. matchmaker-config.json) and have the developer upload it on the Matchmaker page of the dashboard. The matchmaker starts an authoritative dedicated server for each match, which is the recommended setup for multiplayer games.
List the Edgegap agent skills: tested, step-by-step workflows for getting a multiplayer game online on Edgegap, with templates (Dockerfiles, matchmaker configs, Unity C# and Unreal C++ drop-ins) and known pitfalls. Before working on a game project, load the matching skill with edgegap_get_skill: edgegap-ngo for Unity with Netcode for GameObjects, edgegap-fishnet for Unity with FishNet, edgegap-purrnet for Unity with PurrNet, edgegap-fusion for Unity with Photon Fusion 2, edgegap-unity for other Unity projects, edgegap-unreal for Unreal Engine (edgegap-eos-voice to add voice chat to it), edgegap-fusion-simplefps for the Photon Simple FPS sample, edgegap-godot for Godot 4 with a Godot server, edgegap-colyseus for a Colyseus server, edgegap-nakama for a game with a Nakama backend, edgegap-playfab for a game that keeps PlayFab Matchmaking, edgegap-threejs for a Three.js or other browser game with its own Node.js server, edgegap-manus for a browser game built with Manus, edgegap for any other engine or platform questions; it routes to the rest. Also says whether a newer skills pack is on GitHub (reads one version number, no Edgegap API call).
Load one Edgegap agent skill: its SKILL.md (the workflow to follow) by default, or one of its reference documents or templates with file. Load the matching skill before acting on a game project — edgegap-ngo for Unity with Netcode for GameObjects, edgegap-fishnet for Unity with FishNet, edgegap-purrnet for Unity with PurrNet, edgegap-fusion for Unity with Photon Fusion 2, edgegap-unity for other Unity projects, edgegap-unreal for Unreal Engine (edgegap-eos-voice for voice chat), edgegap-fusion-simplefps for the Photon Simple FPS sample, edgegap-godot for Godot 4 with a Godot server, edgegap-colyseus for a Colyseus server, edgegap-nakama for a Nakama backend, edgegap-playfab for PlayFab Matchmaking, edgegap-threejs for a Three.js/browser game with a Node.js server, edgegap-manus for a browser game built with Manus, edgegap otherwise — and fetch the references and templates it points to as you reach them. No API call.
Check whether the Edgegap agent skills installed in the developer's agent are the latest, and update them. This hosted server cannot see the developer's disk: pass installed_versions (each loaded skill's metadata.version from its SKILL.md frontmatter) and it compares them with the latest on GitHub and says how to update. Skill text is never downloaded by this tool. The skills served by edgegap_get_skill need no update here.
概要
edgegap-mcp connects your coding agent to Edgegap, the orchestration platform behind live multiplayer titles at 390+ studios. Instead of breaking off to a dashboard or an API reference mid-task, the agent creates the application, registers your container image, deploys it near your players, waits for it to come up, and hands back the connection address.
It works with any dedicated server that runs in a container. In practice that means Unity, Unreal Engine, and Godot projects, alongside custom Go, Rust, and Node backends. If your build produces a headless server and a Dockerfile, it deploys.
Ten tools, curated rather than generated from the API spec — sixty auto-generated tool descriptions measurably degrade an agent's ability to pick the right one. The set covers the golden path: list and create applications, register versions with CPU, memory and ports, deploy, poll to ready with backoff, pull container logs after a crash, stop cleanly. Errors are written so the agent can correct itself. A failed image pull comes back naming the fields to check rather than an opaque 424.
Two ways to run it. The hosted endpoint at https://mcp.edgegap.dev/mcp needs no install and works as a custom connector in Claude, or over HTTP in Claude Code, Cursor and Codex. The npm package @edgegap/mcp runs locally, holds your API token in process memory only, and adds read-only mode, an application allowlist, and a deployment duration ceiling for unattended agents. Edgegap API tokens are organization-wide, so local mode with guardrails is the right choice for any organization with a live game in it.
Apache-2.0 at github.com/edgegap/edgegap-mcp. Registered in the official MCP registry as dev.edgegap/mcp.
Edgegap can execute on demand any Docker container image, anywhere, within 1 second.
よくある質問
Edgegap の MCP サーバーにはどう接続しますか?
エンドポイント https://mcp.edgegap.dev/mcp を Claude Code、Cursor、VS Code など任意の MCP 対応クライアントに追加してください。このページのセットアップ手順で各クライアントを一度に設定できます。
Edgegap の MCP サーバーは認証が必要ですか?
はい。Edgegap は API キーが必要です——取得方法は公式ドキュメントを確認し、クライアントの設定に追加してください。
Edgegap の MCP サーバーはどのトランスポートを使用しますか?
Edgegap は Streamable HTTP エンドポイントを公開しており、これはリモート MCP サーバーで広く使われ、主要な MCP クライアントがサポートするトランスポートです。
他のリモート MCP サーバー
AIsa
AIsa is one key for 950+ data APIs across SEO and AI visibility, finance, social, web search, sales and agent mail — the go-to-market data stack behind prospecting, competitor research and brand monitoring. Five tools ra
AdaptlyPost
AdaptlyPost: Schedule and publish social posts to Instagram, TikTok, YouTube, X, Facebook, LinkedIn, Pinterest, Threads, and Bluesky from any MCP client.
AdisInsight
Pharmaceutical drug & clinical trial intelligence
Aard
Aard connects Macroeconomic and other official data from 170+ publishers including World Bank, IMF, BIS, ECB, Eurostat, and National Statistical Offices. Aard is powered by a large graph of the metadata universe and an
AWS Marketplace
Discover, evaluate, and buy solutions for the cloud
コメント