Give your AI Agent real fund data
All it takes is one URL — no Python, no scripts, no process to babysit. Once configured, you ask in Claude Desktop / Cursor / VS Code and the agent calls these 19 tools for real data instead of making things up from memory.
https://api.fundapi.cn/mcpAuthorization: Bearer cf-…POST onlySource and self-hosting:GitHub repository(README includes per-client configs, a self-test and the changelog)
Let your agent install it
Skip finding config files and editing JSON. Paste this into your agent — it reads the official setup guide, configures your client and verifies the connection.
Please read the MCP setup guide first: https://fundapi.cn/en/install-mcp-workflow.md Then configure and verify the Fund Data API MCP server. Client: figure it out yourself from the runtime (Claude Code / Cursor / VS Code / CodeBuddy / WorkBuddy / OpenCode …) — do not ask me. API key: use it if I provided one; otherwise verify connectivity with the free tools only and tell me which tools need a key. Need a key: get one at https://fundapi.cn/en/pricing (Alipay QR, issued on the spot). Requirements: 1. Authenticate with the Authorization header — do not switch to OAuth. 2. Do not skip verification: after the handshake, confirm 19 tools are listed. 3. If configuration fails, give the concrete cause and a fix — do not guess.
Prefer doing it yourself? Manual setup below — three lines per client.
Three steps
Get your API key
The free tools (search / profile / screener options)work without a key. Everything else needs a key for the matching tier. It is shown in plain text exactly once, at issue time.
See tiers and pricingAdd it to your client
Paste this into your client's MCP config — change only the key:
{
"mcpServers": {
"fundapi": {
"type": "http",
"url": "https://api.fundapi.cn/mcp",
"headers": {
"Authorization": "Bearer cf-YOUR_KEY"
}
}
}
}| Client | Config file |
|---|---|
| Claude Desktop | claude_desktop_config.json |
| Claude Code | .mcp.json |
| Cursor | .cursor/mcp.json |
| VS Code | .vscode/mcp.json |
| Codex | ~/.codex/config.toml |
| Gemini CLI | ~/.gemini/settings.json |
| Cline | ~/.cline/mcp.json |
| Windsurf | mcp_config.json |
| CodeBuddy | ~/.codebuddy/.mcp.json |
| WorkBuddy | ~/.workbuddy/mcp.json |
| Trae | Settings → MCP → Manual config |
| Kimi Code | ~/.kimi/mcp.json |
| OpenCode | opencode.json |
| Kilo Code | kilo.jsonc |
| ZCode | ~/.zcode/cli/config.json |
| MiMo Code | mimocode.json |
| Manus | Settings → Integrations → Custom MCP |
| Devin | ~/.config/devin/mcp_config.json |
| OpenClaw | ~/.openclaw/openclaw.json |
No key yet? Delete the headers line and it still works — the free tools need no credentials. Add a key when you want all 19.
If your client cannot send custom headers (some web connectors), usehttps://api.fundapi.cn/mcp?key=cf-YOUR_KEY instead — same auth, same quota, same rate limits, only the credential location differs.
Restart your client and just ask
After the restart you will see 19 fund_* tools. Try asking:
- “How is the E Fund Consumer Sector fund doing?”
- “Which stocks did it add last quarter?”
- “How many times has this fund changed managers?”
- “Do these two funds hold the same stocks?”
- “Find me a few five-star consumer funds”
Check the endpoint without a client
One request proves the MCP service is up — no key, nothing to install:
curl -s -X POST https://api.fundapi.cn/mcp \
-H "Content-Type: application/json" \
-H "Accept: application/json, text/event-stream" \
-d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"check","version":"1"}}}'A response containing "serverInfo":{"name":"fund-data-api"means the service is live; the tool list only appears after a client handshake.
19 tools
Tiers match the HTTP endpoints — free tools need no key| Tool | What it does | Tier |
|---|---|---|
fund_search | Find a fund by code, name or pinyin. Use it first when the user only knows the fund name. | Free |
fund_profile | Fees, 10 period returns with peer rank, manager, AUM, allocation, holder structure — in one call. | Free |
fund_screener_options | Screener dictionary: type tree, companies, themes, return buckets, rating agencies, sort fields. | Free |
fund_nav | NAV history: unit NAV, accumulated NAV, daily growth, subscription status, date range. | Trial |
fund_rank | Whole-market ranking by period return, top N. | Trial |
fund_notices | Notice index: periodic reports, company announcements, personnel changes. | Trial |
fund_holdings | Quarterly top 10 holdings: weight in NAV, shares held (10k shares). | Pro |
fund_managers | Current managers: tenure, AUM, overall score plus 5 sub-scores, tenure return vs peers. | Pro |
fund_estimate | Intraday estimate: realtime and holdings-model, with per-stock contribution. | Pro |
fund_dividends | Dividends and share splits with related notices. | Pro |
fund_trade_rule | Trading rules: subscription status, T+N settlement, tiered fees, channel discounts. | Pro |
fund_rating | Agency star ratings by reporting period — see upgrades and downgrades. | Pro |
fund_overlap | Overlap of top holdings across funds — are you buying the same stocks twice? | Institutional |
fund_portfolio_xray | Portfolio look-through: combined holdings of a basket, each stock's weight and which funds it came from. | Institutional |
fund_holding_changes | New entries, dropouts, adds and cuts this quarter, with multi-period trend. | Institutional |
fund_manager_history | Past managers, per-tenure returns, turnover frequency. | Institutional |
fund_portfolio_diagnose | Portfolio diagnostics: industry exposure, look-through, position trend, weighted fees. | Institutional |
fund_screener | Market screener by type, company, theme, return range and rating, with sorting and paging. | Institutional |
fund_compare | 2–5 funds side by side on returns, fees, managers, holdings (juxtaposition only — no ranking, no verdict). | Institutional |
The 19 tools map one-to-one onto the 19 HTTP endpoints. Parameters and field reference live in theAPI reference.
How MCP relates to HTTP
One service, two ways to call it:MCP for your agent, HTTP for your code — both use the same endpoints, the same API key and the same quota, with identical data and billing.
Pick by caller: MCP for agent workflows, HTTP for in-code calls. You can use both at once; MCP needs no separate purchase.
For offline setups or clients limited to local commands, you can also run it over stdio:fund-mcp.py(needs Python 3 + requests; configure FUND_API /FUND_KEY), or clone theGitHub repositoryand self-host it.
Tools not showing up, or the handshake fails? Check theFAQ, orget in touch.