Prompeteer

io.github.prompeteer由 GitHub 帳號 prompeteer 發布v2.0.3更新於 Oct 24, 2026

Contextual prompts and agent skills for 140+ AI platforms.

已驗證Streamable HTTP可網頁執行效率與工作流知識與記憶AI 與機器學習

概覽

AI 產生的概覽

Prompeteer 為 140 多個 AI 平台產生情境化提示詞與代理技能,可評分並存到 PromptDrive,還能搜尋私有 Memory。

可以做什麼
  • 針對實際使用的模型產生提示詞與代理技能,涵蓋多平台與多語言
  • 以 16 個維度為提示詞評分
  • 列出、讀取並儲存 PromptDrive 中的提示詞
  • 搜尋並查詢私有 Memory 來源,讀取或更新 Memory 設定
適用情境
適合需要讓助理為特定模型草擬、評分並保存可重複使用的提示詞或技能,或在支援的 MCP 用戶端中查詢自己的 Prompeteer Memory 時使用。
執行需求
需要 Prompeteer 帳號,並透過 OAuth 2.1 與 PKCE 在瀏覽器登入。可使用託管的 Streamable HTTP 端點,或使用 npm stdio 橋接,後者需 Node.js 22.14.0 以上版本並會開啟瀏覽器授權。需能連線至 Prompeteer 服務及其 OAuth 端點。
安裝前請注意
產生、評分與 Memory 操作可能消耗方案額度,成功產生的結果會自動存到 PromptDrive。儲存與變更 Memory 設定會寫入帳號。工具輸入與所請求的結果由 Prompeteer 服務處理,可能使用第三方 AI 供應商。可用功能取決於同意的權限範圍(mcp:read、mcp:write、mcp:generate、memory:read、memory:write)。產生的技能僅以內容回傳,不會被執行。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 Prompeteer,將其新增到工作區。
  2. 為需要使用其工具的對話啟用該服務。

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。

交給代理程式安裝

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/io.github.prompeteer%2Fprompeteer-mcp/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。

檢視代理程式所依據的安裝指南

其他 MCP 客戶端

把它新增到你客戶端的 mcpServers 設定中。

{
  "mcpServers": {
    "prompeteer": {
      "type": "http",
      "url": "https://prompeteer.ai/mcp"
    }
  }
}

README

Prompeteer MCP Server

Contextual prompts and agent skills for 140+ AI platforms.

Prompeteer generates contextual prompts and agent skills tuned to the model you are actually using, across 140+ AI platforms and 129 languages. Every prompt gets a 16-dimension Prompt Score, and skills get a Skill Score. Saved prompts live in PromptDrive and roam across platforms. Free to start.

Connect to the hosted server

SettingValue
Endpointhttps://prompeteer.ai/mcp
TransportStreamable HTTP
AuthenticationOAuth 2.1 with PKCE S256
Connection guideprompeteer.ai/connect

Use a client that supports remote MCP and OAuth discovery. Add the endpoint in its remote-server settings and complete the browser sign-in. Access depends on the account, plan, consented scopes, and client support.

For Cursor, add this entry to .cursor/mcp.json:

json
{  "mcpServers": {    "prompeteer": {      "url": "https://prompeteer.ai/mcp"    }  }}

Other clients have their own configuration formats. Use the connection guide for client-specific setup.

Connect through the npm stdio bridge

For clients that launch local MCP commands, use Node.js 22.14.0 or later and add:

json
{  "mcpServers": {    "prompeteer": {      "command": "npx",      "args": ["-y", "@prompeteer.ai/[email protected]"]    }  }}

The command runs a stdio bridge to the same hosted Prompeteer endpoint. It uses the pinned mcp-remote 0.8.6 proxy for Streamable HTTP and OAuth. The client opens a browser for sign-in when authorization is needed. The local bridge does not run Prompeteer's generation models or product service.

The endpoint and OAuth resource are fixed to https://prompeteer.ai/mcp. The wrapper uses a loopback OAuth callback, runs without a shell, and forwards MCP messages through stdout. Connection diagnostics use stderr. It does not accept arbitrary endpoints, authentication headers, or debug flags. Native remote clients can use the direct URL without installing the bridge.

sh
npx @prompeteer.ai/[email protected]npx @prompeteer.ai/[email protected] --helpnpx @prompeteer.ai/[email protected] --jsonnpx @prompeteer.ai/[email protected] --version

--json prints the direct remote configuration for Cursor. --help and --version do not start a connection.

OAuth credentials are managed by mcp-remote, using its local ~/.mcp-auth cache by default. On POSIX systems, new credential files are written with mode 0600. The bridge does not add its own token store. The package includes the tested proxy runtime and the patched query parser used by its callback server. Client installation does not resolve a different proxy dependency tree.

Capabilities

Tool familyExamples
Prompt and skill generationgenerate_prompt, generate_skill
Prompt Scorescore_prompt
PromptDrivelist_prompts, get_prompt, save_to_promptdrive
Private Memorylist_memory_sources, search_memory, get_memory_page, ask_memory, get_memory_settings, update_memory_settings

Skills receive a Skill Score as part of the skill workflow; score_skill is not a separate advertised MCP tool. Discover the current tool schemas through the authenticated server. Generation, scoring, writes, and private Memory operations may require additional scopes or consume plan allowances.

Example: “Generate a prompt for a product launch email in Claude. Show its Prompt Score, then save the approved version to PromptDrive.”

Authentication

The server advertises OAuth metadata and requests the scopes needed for authorized work. Keep credentials in the MCP client's managed authentication flow.

ScopePurpose
mcp:readRead prompts and vault metadata
mcp:writeSave prompts
mcp:generateGenerate prompts and skills
memory:readRead private Memory
memory:writeUpdate Memory settings

OAuth metadata and protected resource metadata are publicly discoverable. An unauthenticated MCP request returns HTTP 401 until you sign in.

Pricing and public facts

Free to start. See current plans and allowances.

Prompeteer reports 12,000+ users across 160 countries and supports 129 languages. These company facts were approved for publication in September 2026.

Registry and source

The official MCP registry lists the server as io.github.prompeteer/prompeteer-mcp. This is separate from first-party connector directories.

server.json describes both the npm stdio bridge and the hosted Streamable HTTP endpoint. Its registry description uses the canonical tagline because the registry limits descriptions to 100 characters. The full description is preserved in publisher metadata. .mcp/server.json is an identical compatibility copy.

Version 2.0.3 adds a working stdio bridge and corrects stale setup instructions. Earlier npm 2.0.0 printed setup instructions and exited, even though registry versions 2.0.0 through 2.0.2 advertised it as a stdio transport. The hosted endpoint remains https://prompeteer.ai/mcp.

Repository manifests for Cursor and Gemini CLI

plugin.json and mcp.json provide the portable Agent Plugins format supported by Cursor. After installing the plugin through Cursor, connect Prompeteer and complete browser sign-in. The direct Cursor configuration above also works as a manual setup format. Generated skills are returned as content; connecting does not run or install a generated skill.

gemini-extension.json provides a Gemini CLI extension using the hosted HTTP endpoint and OAuth discovery. Once the manifest is available in the public repository, install with:

sh
gemini extensions install https://github.com/prompeteer/prompeteer-mcp

Restart Gemini CLI and complete the Prompeteer connection when prompted. Generation and scoring can consume account quota; requested saves write to your PromptDrive.

These manifests have been structurally validated. Installation and OAuth in the actual Cursor and Gemini CLI applications have not been tested for this release. Their presence does not establish a marketplace listing or platform approval. See the Cursor plugin documentation and Gemini extension reference.

Plugin source for Claude Code, Cowork and Grok Build

.claude-plugin/plugin.json supplies plugin metadata and .mcp.json connects the client directly to https://prompeteer.ai/mcp over HTTP. This plugin contains one remote MCP connection. It adds no local command, hook, skill, agent, or shell execution. It does not start the npm bridge or install its runtime dependencies.

For a local Claude Code check after cloning this repository:

sh
claude plugin validate .claude --plugin-dir .

Use /mcp in Claude Code to complete the Prompeteer OAuth connection. For Grok Build, the repository can be installed through a marketplace entry that pins its exact public commit. Marketplace submission, listing, and functional client testing are separate outcomes; these source files alone do not establish any of them.

Network access and account permissions

The client contacts https://prompeteer.ai/mcp and Prompeteer's OAuth discovery, registration, authorization and token endpoints on https://prompeteer.ai. Prompeteer sign-in may redirect to the identity provider selected by the user. Credentials remain in the client's supported OAuth flow; the plugin contains no tokens, API keys or environment-variable readers.

A Prompeteer account is required. Tool availability depends on the authenticated account, approved scopes and plan allowances. PromptDrive tools read or save account records; Memory tools access the user's Prompeteer Memory when enabled and authorized. Memory refers to Prompeteer's stored sources, not the host assistant's memory or conversation history. Generation and scoring can consume quota, and successful generation automatically saves its result to PromptDrive as part of the service workflow. The actual returned savedToVault and promptId fields indicate the persistence outcome. Generated skills are returned as content; this plugin does not execute them.

Privacy policy

Tool inputs and any requested saved-content or Memory results are processed by the Prompeteer service and returned to the connected client. Prompeteer's backend may use AI providers to carry out generation or scoring. See the published privacy policy for service data practices, terms, and plans. Contact [email protected] for support or privacy questions. Disconnect the MCP connection in the client when it is no longer needed.

Build and verify from source

Use Node.js 22.14.0 or later. The source lock pins the maintained proxy and its dependencies. Build copies those unmodified runtime files, with their licenses, into the release package.

sh
npm ci --ignore-scriptsnpm run buildnpm run checknpm testnpm auditnpm pack --dry-run --json --ignore-scripts

The explicit build uses mcp-remote 0.8.6 and patched qs 6.16.0. Tests run protocol forwarding and startup-disconnect checks against local fixtures; they do not generate prompts or make paid service calls. The vendor/ build output is included in npm releases and is not committed to this repository.

Support and license

The bridge wrapper and documentation in this repository are MIT licensed. The hosted Prompeteer platform is proprietary.

來源:README.md,提交 e035a0e

工具

0
SourceWeft 尚未讀取此伺服器的工具。

版本歷史

1
  1. v2.0.3最新Sep 16, 2026

評分

已滿足適用檢查項的 85%

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。

  • 近 30 天內可連通必需25 分
  • 已宣告認證方式必需10 分
  • 已列出工具必需15 分

    請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。

  • 描述必需15 分
  • 圖示10 分
  • 文件15 分
  • 授權條款10 分