dev.fiskmas由 fiskmas.dev 發布官方發布者v1.0.0更新於 Oct 24, 2026
Hosting for AI agents: your AI client deploys Docker apps to live HTTPS URLs over MCP.
Fiskmas 是遠端 MCP 伺服器，讓 AI 助理在 Fiskmas 主機服務上建立、部署、執行與監控 Docker 專案，並取得可公開存取的 HTTPS 網址。
Web executable，透過 Streamable HTTP。 遠端服務在工作區中設定後即可從網頁執行環境執行。
把這段提示詞貼上到 Claude Code、Codex、Cursor 等代理程式，或 SourceWeft 對話中。代理程式會讀取該伺服器的安裝指南，先向你展示發布者、端點和登入方式，經你同意後再把它新增到你的客戶端。金鑰不會寫進任何命令。
閱讀 https://sourceweft.com/mcp/dev.fiskmas%2Ffiskmas/install.md，並按其中說明把這個 MCP 伺服器接入我的客戶端。修改任何設定之前，先告訴我它的發布者、端點和登入方式，等我確認。不要把 API 金鑰寫進命令或我會提交的檔案；問我金鑰應從哪裡取得。把它新增到你客戶端的 mcpServers 設定中。
{
"mcpServers": {
"fiskmas": {
"type": "http",
"url": "https://mcp.fiskmas.dev/mcp",
"headers": {
"Authorization": "Bearer <YOUR_TOKEN>"
}
}
}
}Return the identity of the current token holder (account + provider) plus its subscription plan (free | plus) and the plan's limits: max running projects, per-project memory/CPU caps, log windows, and the project-database fair-use guide. Free plans include an upgrade_url for Plus ($9.99/mo: 5 projects, 512 MB / 1 CPU per project, always-on).
無需輸入參數。
Start onboarding: require an email, create a pending account, and email a human verify link. Does NOT return an API token. The human must click the link; the verify page shows the token once. Then configure Authorization: Bearer <token> on this MCP client (also docker login -u <token> registry.fiskmas.dev). If you already have a token, call whoami instead.
email必填display_namePoll onboarding state for an email (pending | active | unknown). Use after create_account while waiting for the human to verify.
email必填Re-issue the API token for an existing account. Provide the email the account was registered with; Fiskmås emails a fresh one-time verify link to that address. The human clicks it; the verify page shows the new token once (old tokens are revoked immediately). This is the self-service recovery path when a token was lost, rotated, or is being rejected as "invalid or expired". It does not create a new account — for a brand-new account use create_account. Rate-limited: a few requests per hour per email.
email必填Delete the caller's account. Stops and removes all projects (containers, routes, MariaDB) first, then deletes the account. Refuses to leave orphaned runtimes.
無需輸入參數。
Create a project and reserve <slug>-app.fiskmas.dev. Slug must match ^[a-z0-9][a-z0-9-]{1,48}$ and is not reserved. Subject to the plan's running-project quota (Free: 1, Plus: 5) — on quota the error tells the human how to upgrade or free a slot.
nameslugList the caller's projects with status and URL.
無需輸入參數。
Fetch a single project by id or slug.
projectStop and delete a project. Its slug goes into quarantine for 7 days: you may re-use it immediately, other accounts get a 409 until it is freed.
projectSet environment variables on a project. They take effect on the next deploy_project: a restart does NOT re-apply env.
envprojectList a project's env keys with masked values (never returns full secrets).
projectPull an image by digest and run one container with the plan's cpu/memory caps (Free: 256 MB / 0.2 CPU, Plus: 512 MB / 1 CPU). Only registry.fiskmas.dev images are accepted. Rate-limited to 1 deploy per 5 minutes per account (429 with a retry-seconds hint); use restart_project for a transient crash - it does not count against the limit. Re-deploying a running project is always allowed; starting a second one on Free is blocked until one stops.
imageprojectStart a stopped project's container. Waits until it is ready; a pod that fails to become ready returns a named error.
projectStop a running project's container (keep the image).
projectRestart a running project's container. Recreates the pod, re-running the injected-DB resolver gate and the app's storage probes, and waits for readiness: a failed start returns a named error (e.g. an unresolvable DB host) and a plain retry usually clears a transient DNS drop. Not rate-limited. Env changes are NOT picked up by a restart - redeploy instead.
projectReturn project status, url, image digest, and last deploy error.
projectReturn recent container log lines. `tail` = line count (plan default 200; plan max 2000), `since` = time window in seconds (plan window: free 24 h, plus 7 d — logs are live-pod streams, not stored history). When both are given, `since` wins.
tailsinceprojectAccount-wide or per-project stats for debugging: live CPU/mem vs the plan caps (Free: 256 MB / 0.2 CPU, Plus: 512 MB / 1 CPU) and deploy history. Omit `project` for all of the caller's projects; pass a project id or slug NaNalso returns `resources.history` — per-project CPU/mem time series over the window (sampled every ~30 s on the k3s platform; absent on stacks without the metrics collector).
sinceprojectReport a problem with the Fiskmås toolset so it can be fixed. Use it when a tool errored or misbehaved, you got stuck and can't proceed, or the instructions/skill were unclear. Do NOT use it for praise. kind=bug means it broke, kind=suggestion means it was unclear or could be better, kind=other is anything else that didn't work. Name the `tool` and, if relevant, the `project`. At most one call per session.
kind必填toolmessage必填projectList the platform skills available to the caller.
無需輸入參數。
Return the bytes of a platform skill by name.
name已滿足適用檢查項的 83%
該評分衡量此收錄條目的完整程度，依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價，也不表示伺服器安全或目前可用：可連通性是帶日期的檢查結果，評分也不包含安裝量或使用者評價。
請宣告包含 README 的 GitHub 儲存庫，或在 OAuth 受保護資源後設資料中提供 resource_documentation。
託管服務按其提供方的條款使用，因此這裡不要求授權條款。此檢查項不計入百分比。