Zyfai

io.github.ondefyv1.0.0更新於 Oct 3, 2026

Discover DeFi yield opportunities and manage Zyfai positions with self-custodial AI agents.

已驗證Streamable HTTP可網頁執行AI & MLFinance

概覽

AI 產生的概覽

讓助理探索 DeFi 收益機會，並在通過驗證後查看 Zyfai 持倉、準備支援的帳戶操作。

功能
這是一個託管於遠端的 Streamable HTTP MCP 伺服器，透過自然語言工具提供 Zyfai 的收益探索與 DeFi 代理能力。它能列出保守型與積極型收益機會，並比較 APY、TVL、協議、池子、鏈、資產與風險資訊。在已通過驗證的 Zyfai 工作階段中，還能查看投資組合與目前持倉、檢視收益、交易歷史與再平衡資訊，並在啟用時準備支援的存款與帳戶操作。
適用情境
適合讓助理研究或比較 DeFi 收益策略，或檢視並準備現有 Zyfai 帳戶上的操作，而不必自行建置整合。公開的機會探索不需登入；投資組合與寫入流程則需要驗證。
執行需求
需要支援 Streamable HTTP 的遠端 MCP 用戶端，並指向託管端點。使用託管伺服器不需本機執行環境；若改為本機執行，則需要 Node.js 18+ 與 pnpm，並在 .env 中設定 ZYFAI_API_KEY 才能呼叫即時 API。個人投資組合與寫入工具需要已驗證的 Zyfai 工作階段，OAuth 挑戰會透過 HTTP 401 與 WWW-Authenticate 回傳。
安裝前請注意
投資組合、持倉、收益、歷史與個人化工具需要驗證，支援的寫入流程可能回傳 Zyfai 簽名網址。連接此伺服器不會轉移資產，也不會讓助理託管錢包，且它從不需要在對話中提供助記詞或原始私鑰。簽名前請確認鏈、資產、金額、策略、目的位址與錢包請求。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 Zyfai，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。

其他 MCP 客戶端

把它新增到你客戶端的 mcpServers 設定中。

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

README

Zyfai DeFi MCP Server

Use Zyfai from Claude, Grok and Grok Bot, Cursor, Claude Code, compatible custom agents, and a limited ChatGPT preview.

This public Model Context Protocol server exposes Zyfai's yield discovery and personalized DeFi agent through natural-language tools. Discover opportunities, compare strategies, inspect positions, review earnings, and prepare supported account actions without building a separate integration.

Hosted MCP endpoint: https://mcp.zyf.ai/mcp

You do not need to clone or run this repository to use the hosted server.

What you can do

  • Discover conservative and aggressive yield opportunities
  • Compare APY, TVL, protocol, pool, chain, asset, and available risk information
  • Inspect an authenticated user's Zyfai portfolio and active positions
  • Review earnings, transaction history, and rebalancing information
  • Prepare supported deposits and account actions when enabled
  • Integrate Zyfai into any compatible Streamable HTTP MCP client

The server is built on @zyfai/sdk and uses remote Streamable HTTP transport.

Connect a client

ClientHow to connectPositioning
ClaudeAdd a custom connector using the hosted endpointRecommended conversational experience
CursorAdd the JSON configuration belowDeveloper workflows
Claude CodeRun the command belowTerminal and coding-agent workflows
Grok and Grok BotAdd a Custom connector at grok.com/connectorsLonger-running and repeat agent workflows
ChatGPTLimited private preview; currently desktop-onlyPublic distribution not yet available
Other clientsConfigure the hosted URL as a remote Streamable HTTP serverCustom agents and integrations

For current screenshots, authentication behaviour, mobile/desktop availability, and troubleshooting, see Use Zyfai with AI.

Cursor

json
{  "mcpServers": {    "zyfai": {      "url": "https://mcp.zyf.ai/mcp",      "transport": "http"    }  }}

Claude Code

Run this outside an active Claude Code session:

bash
claude mcp add --transport http zyfai https://mcp.zyf.ai/mcp

Claude

In Claude, open Settings → Connectors, add a custom connector, and use:

text
https://mcp.zyf.ai/mcp

Grok and Grok Bot

Open grok.com/connectors, select New Connector → Custom, and enter the hosted endpoint. In Grok Bot, attach the installed Zyfai connector with @ when needed.

Grok Bot is the better fit for longer-running or repeat workflows. Grok web also supports the connector for ad hoc conversations, although tool-heavy requests can take longer to complete.

Try it

Start with public opportunity discovery:

Using Zyfai, show me conservative USDC opportunities on Base. Compare APY, TVL, protocol, and risk.

Then compare strategies:

Now compare those with the aggressive strategy and explain the additional risk rather than recommending purely by APY.

After authentication:

Show my current Zyfai positions and earnings. Do not prepare any transaction.

For a controlled action flow:

Help me prepare a 1 USDC deposit on Base using the conservative strategy. Explain each step and ask before anything requiring a wallet signature.

Authentication and safety

Zyfai separates general discovery from personal account access.

  • Opportunity and protocol discovery use public market and strategy data.
  • Portfolio, position, earnings, history, and personalized tools require an authenticated Zyfai session.
  • Supported write flows require authentication and may return a Zyfai signing URL.
  • Connecting the MCP server does not transfer assets or give an AI custody of a wallet.
  • The server never needs a seed phrase or raw private key from chat.
  • Users should verify the chain, asset, amount, strategy, destination, and wallet request before signing.

OAuth challenges are returned through standard HTTP 401 and WWW-Authenticate responses so compatible clients can begin the authentication flow.

Documentation

AudienceStart here
AI assistant usersUse Zyfai with AI
Application and agent developersZyfai SDK quickstart
SDK-based autonomous agentsAgent quickstart
MCP server contributorsAGENTS.md

Run locally

Requirements: Node.js 18+ and pnpm.

bash
git clone https://github.com/ondefy/zyfai-mcp-servercd zyfai-mcp-servercp .env.example .envpnpm installpnpm run checkpnpm start

Local endpoints:

  • MCP: http://localhost:3005/mcp
  • Health: http://localhost:3005/health

Set ZYFAI_API_KEY in .env for live Zyfai API calls. See .env.example for authentication, backend, CORS, write-tool, and signing configuration.

For workspace-linked SDK development, local OAuth, integration testing, and the full repository architecture, read AGENTS.md.

Validate changes

bash
pnpm run check

This runs TypeScript validation, the production build, and unit tests. Integration suites are opt-in; see AGENTS.md.

Related projects

This public server is separate from zyf-knowledge-mcp, Zyfai's internal code and documentation search service.

License

ISC

來源：README.md，提交 a533d4e

工具

0
工具後設資料尚未被收錄。

版本歷史

1
  1. v1.0.0最新Oct 3, 2026