Context7
Up-to-date docs for AI agents
作者提供的说明(英文原文)
Search and manage Hamster briefs, tasks, plans, and knowledge; trigger delivery from your agent.
这三项永远不会合并成一个分数:意见、测量和人气是不同的东西。
我们还没有为它写通俗的介绍。下面是我们从注册中心读取到的事实。
选择你使用的 AI 应用,然后复制设置。
你需要
你用的是哪个应用?
在上方选择你的应用。我们会记住它,下次按钮会显示“安装到 …”。
适用于网页版、桌面版和手机版 Claude:
https://tryhamster.com/mcp在终端中运行这条命令
claude mcp add --transport http hamster https://tryhamster.com/mcp粘贴到项目中的 .cursor/mcp.json,或 ~/.cursor/mcp.json
{
"mcpServers": {
"hamster": {
"url": "https://tryhamster.com/mcp"
}
}
}浏览器会先询问是否打开应用,然后应用会显示设置并请你确认。密钥的值会留空,由你自己填写。
粘贴到项目中的 .vscode/mcp.json
{
"servers": {
"hamster": {
"type": "http",
"url": "https://tryhamster.com/mcp"
}
}
}浏览器会先询问是否打开应用,然后应用会显示设置并请你确认。密钥的值会留空,由你自己填写。
ChatGPT 只能连接互联网上的服务器:
https://tryhamster.com/mcp菜单名称可能因套餐和版本不同而略有差异。
你的选择通过 Cookie 保存在这台设备上,无需账户。
这些设置是根据作者在注册中心的条目生成的。如果 README 的写法不同,请以 README 为准。
我们还没有读取这个服务器的工具列表;我们会轮流检查各个端点。
我们检查的是任何人都能看到的信息;我们没有阅读代码。不适用的检查项不计入总数。 · 检查于 2026年11月7日
注册名称已通过 GitHub 账户 gethamster 验证,该账户拥有此仓库
我们还没有从 GitHub 读取这个仓库
我们还没有从 GitHub 读取这个仓库
尚未检查;我们会轮流检查各个端点
尚未检查;我们会轮流检查各个端点
尚未检查;我们会在检查端点时读取工具列表
与注册中心里这个分类中所有活跃的服务器比较。只有当发布者拥有该仓库时,星标数才计入。
它在 1 项适用的基础检查中通过了 1 项。我们没有阅读代码。在添加密钥或令牌之前,请先看看这个服务器要求哪些权限,并使用只具备所需权限的密钥。
注册中心没有列出必需的密钥,不过有些服务器会在首次使用时要求你登录。
它在作者的云端运行。通过互联网连接,无需安装,但你发送的内容会经过作者的服务器。
GitHub 仓库最后更新于–,最新版本(0.1.1)于18天前进入注册中心。
支持 MCP 的应用,例如 Claude Desktop、Claude Code、Cursor 和 VS Code。它通过 URL 连接,所以接受远程 MCP 服务器的应用也可以使用,例如开启开发者模式的 ChatGPT。
同一分类中有:Context7, Codebase Memory, Exa
同一分类中其他作者的服务器,按星标数排序。
如何看卡片:★ GitHub 星标i三大支柱:评价 · 检查 · 使用情况i更新 = 最新版本发布到注册中心的时间i
Up-to-date docs for AI agents
Codebase knowledge graph for AI agents — 162 languages, sub-ms queries, 99% fewer tokens.
Connect AI agents to Exa for web search, content fetching, and multi-step research.
按原文从作者的仓库复制。PopNx MCP 没有审查或背书其内容。HTML、图片和脚本已被移除。链接会离开本站。
Install Hamster. One plugin. Talk through hosted MCP or the CLI. Keep the plan on disk, then ship from it.
This is the Hamster product: Agent Plugins skills on every supported client, hosted MCP on every client that supports MCP and installs from GitHub or a marketplace, plus generated native execution workers on Claude Code, GitHub Copilot CLI, and Grok Build. The CLI is how the plan stays in this repo. It is not a second install.
Installation differs by client. If you use more than one, install Hamster separately for each one.
https://github.com/gethamster/plugin at User scope and select Import.On Enterprise, an admin must allow marketplace imports. Once Hamster is on the Cursor marketplace, /add-plugin hamster works too:
/add-plugin hamster
Grok Bot is not a separate Hamster package. It uses the same Cursor account and plugin library, so the Cursor install above is the Grok Bot install. Do not run /add-plugin in the Grok Bot chat. If Grok Bot shows "Complete GitHub auth to sync installed plugins", complete GitHub auth when Grok Bot asks.
/plugin marketplace add gethamster/plugin
/plugin install hamster@hamster-plugins
codex plugin marketplace add gethamster/plugin
codex plugin add hamster@hamster-plugins
You can also launch codex, run /plugins, and install hamster@hamster-plugins.
The Hamster listing in OpenAI's Plugins Directory, used by Codex and ChatGPT, is a zip of this same plugin produced by node scripts/build-codex-bundle.mjs: the Codex manifest, the hosted MCP connector in .mcp.json, the skills, and the listing images. CI builds it on every push and uploads it as a workflow artifact, and the artifact download is the zip to upload to the directory.
One section differs. The zip replaces the "Install the CLI" section of skills/setup/SKILL.md with codex-directory/install-cli.md: when the CLI is missing, setup asks you to install it yourself from CLI binary instead of running the installer, then signs you in and syncs as usual.
Hamster connects to your workspace through the hosted MCP server. Coding skills require access to your repository and a shell.
From a clone of this repository:
agy plugin install ./plugins/hamster
Or from GitHub:
agy plugin install https://github.com/gethamster/plugin/plugins/hamster
Both install plugins/hamster, the folder every other client installs. The bare repository URL also works, but agy then imports the root as a Gemini CLI extension and drops the agents' model settings; running the command above replaces that install in place.
If Hamster 3.2 is already installed, uninstall it before you install 3.4. An in-place Antigravity install retains the four skill directories renamed in 3.4.
agy plugin uninstall hamster
agy plugin install https://github.com/gethamster/plugin/plugins/hamster
In the Antigravity app (2.15 or newer), authenticate hamster_hamster under Settings → Customizations → Installed MCP Servers. The plugin registers that server itself, so a manual hamster entry in mcp_config.json would add a second copy of every tool.
Using GitHub Copilot CLI, Gemini CLI, Grok Build, OpenCode, or Pi? See More install options.
copilot plugin marketplace add gethamster/plugin
copilot plugin install hamster@hamster-plugins
To sign in to the hosted MCP server, run /mcp auth hamster.
gemini extensions install https://github.com/gethamster/plugin
To sign in to the hosted MCP server, run /mcp auth hamster.
grok plugin install gethamster/plugin --trust
To sign in to the hosted MCP server, open /mcps, select hamster, and press i.
OpenCode takes the skills and the hosted MCP server in two steps. Run these in your project folder:
npx skills add gethamster/plugin --agent opencode
opencode mcp add hamster --url https://tryhamster.com/mcp
Add --global to npx skills add to install into ~/.agents/skills/ instead of the current project (opencode mcp add --url needs OpenCode 1.17.0 or later).
To sign in to the hosted MCP server, run opencode mcp auth hamster.
pi install npm:@tryhamster/pi
To track the repository instead of npm releases, run pi install git:github.com/gethamster/plugin.
The Pi package installs the skills only. To connect the hosted MCP server, add it and sign in:
pi mcp add hamster --url https://tryhamster.com/mcp
pi mcp login hamster
https://tryhamster.com/mcp, or hamster chat when MCP tools are unavailable and the CLI is signed in. Your client owns the Hamster sign-in.hamster auth login, and syncs the plan.Claude Code and Copilot CLI list these as /hamster: . Cursor and Grok Build list them as / , and Pi as /skill: . In OpenCode, ask for one by name, or in OpenCode 2 mention it as @setup. The table uses the Claude form.
| Skill | Persona | Description |
|---|---|---|
/hamster:setup | — | Install the CLI, sign in, and sync the plan into this repo |
/hamster:ask-hamster [request] | Workspace Copilot | Connect current code with workspace priorities, blockers, blueprints, or related work (hosted MCP when the client has it; hamster chat otherwise) |
/hamster:ship [brief] | Release Engineer | Ship a brief: merge base, implement in parallel, test, review, bisectable commits, PR |
/hamster:plan-hamster [brief] | Tech Lead + CEO/Eng modes | Analyze brief with optional founder or architecture review |
/hamster:resume-hamster [brief] | — | Resume interrupted execution from where you left off |
/hamster:review-hamster | Staff Engineer | Paranoid two-pass code review (CRITICAL then INFORMATIONAL) |
/hamster:qa [mode] | QA Lead | Systematic testing: diff-aware, full, quick, regression |
/hamster:retro [days] | Eng Manager | Engineering retrospective with metrics, trends, team analysis |
Four skills carry a -hamster suffix because Cursor invokes plugin skills as a bare /skill-name, and ask, plan, and resume are Cursor's own concepts — cursor-agent --mode takes plan and ask, --resume selects a session — so short names compete with them there. review is renamed with that family so Cursor's command list stays one convention. Claude Code namespaces plugin skills as / : and they cannot conflict, so on Claude the suffix is redundant and you type /hamster:ask-hamster. One skills tree serves every client, so that is the cost of being unambiguous on Cursor. ship, qa, retro, and setup shadow nothing and stay short.
/hamster:setupThe readiness path. Noninteractive check first (ensure-ready). If the CLI is installed and you are signed in, it runs hamster sync to refresh the plan. Otherwise it installs the CLI, opens login, and inits/syncs — only when you asked.
/hamster:ask-hamsterThe gateway for open-ended questions that need Hamster's judgment across the workspace, such as how the code you're changing relates to priorities, blueprints, or decisions. Asks through the ask_hamster MCP tool when the client has it; otherwise hamster chat is the same ask path over the CLI. Direct reads and writes go to the named Hamster MCP tools instead. For example:
/hamster:ask-hamster I'm modifying auth middleware in apps/web/app/api/. What does our blueprint say about third-party integrations?
/hamster:ask-hamster I prototyped rate limiting in apps/api/middleware/rate-limit.ts. Which brief or initiative should this work belong to?
Follow-up questions continue the same Hamster conversation when they depend on the previous response.
/hamster:shipThe main orchestrator. Ship, plan-hamster, and resume-hamster share brief selection: a slug, UUID, or Hamster Studio URL takes precedence, followed by an exact title or short title (the part before a subtitle separator). Names ignore case, a leading “the”, a trailing “brief”, and trailing punctuation. A unique match proceeds without a selection question; ambiguous or partial-only names require clarification.
/hamster:ship user-authentication
/hamster:ship the Checkout reliability brief. /hamster:ship https://tryhamster.com/home/hamster/briefs/2de8d546-50ab-4dbd-a678-579ec8119f60
If no argument is given, presents an interactive picker of actionable briefs.
Flow: Readiness (setup/ensure-ready) → Setup (prereqs + live sync) → Brief selection → Inline wave scheduling (one confirmation) → Branch + merge base → Parallel wave execution (implement → validate + test → wave review → bisectable commits) → Final validation → Ask about PR creation
No plan generation or task elaboration occurs at any step — scheduling only organizes the pre-generated tasks into parallel waves.
/hamster:plan-hamsterRead-only analysis with optional deep review. Produces the execution plan without making changes.
/hamster:plan-hamster api-rate-limiting
After analysis, choose a review mode:
/hamster:resume-hamsterResumes an interrupted execution. Auto-detects the brief from the git branch name (feature/{key}-{id}-{slug}), in-progress tasks, or a provided argument.
/hamster:resume-hamster
/hamster:resume-hamster user-authentication
/hamster:review-hamsterParanoid two-pass code review for the current feature branch:
/hamster:review-hamster
/hamster:qaSystematic testing with 4 modes:
/hamster:qa diff # Test only what changed (default on feature branches)
/hamster:qa full # Full test suite with coverage /hamster:qa quick # 30-second lint + typecheck + smoke tests /hamster:qa regression # Changed files + dependents, flag new failures
Includes issue taxonomy (functional/type-safety/integration/performance/coverage-gap) and optional fix loop.
/hamster:retroEngineering retrospective from git history:
/hamster:retro # Last 7 days (default)
/hamster:retro 14 # Last 14 days /hamster:retro 30 # Last 30 days /hamster:retro 24h # Last 24 hours
Produces: metrics table, hourly distribution, session analysis, hotspots, PR sizes, per-contributor deep dive with praise and growth suggestions, trends vs last retro, and a narrative summary.
| Worker | Persona | Purpose |
|---|---|---|
| task-executor | Senior Engineer | Implements one parent task + subtasks; loads project skills, blueprints, and methods |
| wave-reviewer | Staff Engineer | Reviews a whole wave's diff (per-parent verdicts + cross-parent integration checks), then simplifies |
Canonical worker protocols live in plugins/hamster/skills/ship/references/agents/. plugins/hamster/agents/task-executor.md and plugins/hamster/agents/wave-reviewer.md are generated native adapters (registratio
只显示了开头部分;其余内容请在仓库中阅读