
Agentprofile
io.github.everyai-com由 GitHub 帳號 everyai-com 發布v0.1.0更新於 Oct 27, 2026
One profile — skills, credentials, and memory — synced to every agent tool via one MCP URL.
概覽
遠端 MCP 伺服器,透過單一 URL 為多個 AI 工具提供共用的技能與記憶設定檔。
- 可以做什麼
- 取得目前設定檔的上下文、技能與已存事實
- 列出、讀取並新增技能到共用設定檔
- 儲存、回想與刪除記憶,讓事實跨工具保留
- 在 Claude Code、Cursor、Codex 等 MCP 用戶端之間同步同一份設定檔
- 在網頁儀表板即時檢視與管理記憶和技能
- 適用情境
- 適合同時使用多個 AI 程式開發工具、希望偏好、事實與技能自動延續而不必反覆貼上的人。也適合想要輕量、可自架記憶層而非完整知識庫的使用者。
- 執行需求
- 需要遠端 Streamable HTTP 端點,無須本機安裝。必須驗證:OAuth 自動探索,或在 Authorization 標頭提供 bearer token(可用 CLI 產生)。託管執行個體可建立免註冊的匿名設定檔。自架需要 Cloudflare 帳號及 Workers 與 R2。
安裝
在 SourceWeft 中
- 開啟 儀表板中的 Agentprofile,將其新增到工作區。
- 連接 所需的 憑證。它們會被私密儲存,絕不會在公開頁面上展示。
- 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.everyai-com%2Fagentprofile/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。其他 MCP 客戶端
把它新增到你客戶端的 mcpServers 設定中。 客戶端會在首次連接時完成 OAuth 登入。
{
"mcpServers": {
"agentprofile": {
"type": "http",
"url": "https://agentprofile.everyai-com.workers.dev/mcp"
}
}
}README
agentprofile
[license] [MCP] [clients] [status]
Your agent's identity, everywhere. One profile — skills, credentials, and memory — synced to every agent tool through a single MCP URL. Open source, zero-knowledge, and self-hostable on your own Cloudflare account.
Every AI tool you use — Claude Code, Cursor, Codex — starts from zero: its own skills, its own pasted API keys, its own amnesia. agentprofile gives them one shared brain.
Live now: hosted at https://agentprofile.everyai-com.workers.dev · dashboard at
/app
Connect (any MCP client)
The server is a remote MCP server (Streamable HTTP) — no install needed, just one URL:
It supports two ways to authenticate:
- OAuth (recommended) — clients that support MCP OAuth discover it automatically from the URL and show a consent screen.
- Bearer token — run
npx @magicteams_ai/agentprofileto get a token, or hitPOST /profiles.
Claude Code
Or with a bearer token:
Cursor / other mcp.json clients
Codex CLI
Or, for a client that does OAuth discovery:
VS Code / Copilot
Or just run npx @magicteams_ai/agentprofile and it configures every client it finds.
60-second demo
That creates an anonymous profile (no signup) on the hosted instance, auto-detects your installed MCP clients, and writes their config. Restart your tools, then:
- In Claude Code: "Remember that I prefer pnpm over npm."
- In Cursor: "What package manager do I prefer?"
It recalls it. That's your profile, everywhere.
What's in this repo (Phase 1)
This is the Phase 1 implementation from the master plan: a working end-to-end skeleton that syncs skills and memory across tools over MCP. Credentials (zero-knowledge), OAuth, semantic recall, and the skill registry land in later phases — see Roadmap.
Web UI
/— landing page with the one-command install and the pitch./app— live dashboard. Paste your token (or hit "Create new") and watch skills and memory update in real time: when any agent callsremember, the fact appears instantly over a WebSocket. Add or delete memories and skills right from the page./setup-prompt— the "set up with your agent" prompt: paste it into Claude Code or Cursor and your own agent installs and verifies agentprofile for you.
How it works
- One Durable Object per user is the source of truth. It validates the profile's secret on every call and owns that profile's memory and skills.
- Bearer token =
ap_<profileId>.<secret>. TheprofileIdroutes to the DO; the DO verifies a SHA-256 hash of the secret. No central token table. - The MCP surface is 6 tools:
get_context,list_skills,get_skill,remember,recall,forget. Deliberately small — every description byte is paid on every client turn (pinned in tests).
Run it locally
In another terminal:
Run the tests and typecheck:
Deploy your own (self-host)
agentprofile self-hosts on your own Cloudflare account — the free Workers tier covers personal use.
Roadmap
- P1: MCP server + ProfileDO, 6 tools, memory + skills, instant-profile CLI, live dashboard. ✅
- P2: per-client grants + audit log with live enforcement + dashboard access panel ✅; MCP sessions (
Mcp-Session-Id) ✅; semantic recall via Workers AI embeddings (keyword fallback) ✅; pluggable memory providers (builtin default + Mem0 + Supermemory adapters,/memory-config) ✅; OAuth 2.0 + PKCE consent flow with discovery, dynamic registration, and consent screen ✅. - P3: zero-knowledge credentials (client-side E2E), JIT injection, approval flow.
- P4: one-click Deploy-to-Cloudflare, signed skill registry, format translators.
- P5: teams, SSO, hosted cloud.
See the full plan in docs/.
Contributing
Contributions are welcome under the DCO —
sign off commits with git commit -s. See SECURITY.md for
responsible disclosure.
License
Part of the everyai-com agent stack
Open-source infrastructure for local-first, governed AI agents:
- distillory — local-first memory engine that reasons at ingestion
- agent-ready — turn any app into an MCP server + API + CLI, safe by default
- primer — live business context injected into any agent
- plainsync — local-first Markdown workspace for humans + agents
- argus — cloud-native software verification with parallel browser testing
- mintly-alternative — self-hostable documentation layer for humans + AI agents
- talltrack — sales calls in, publishable content out
Built by Phanindra Reddy · magicteams.ai
來源:README.md,提交 d1909ab
工具
0版本歷史
1- v0.1.0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