Fund data inside DeepSeek Harness
Ask "how has 110022 performed over the past year" or "what was added last quarter" and it calls a tool for real data. The plugin itself is configuration only — data, computation and rate limiting live on the server, so DSH upgrades usually need no change here.
fundapi-dsh-pluginmcp__fundapi__*Endpoint: https://api.fundapi.cn/mcpKey: FUND_KEYSource and changelog:GitHub·npm
Install
Desktop app / Web UI
Open Plugins in the sidebar and enter the package name. Restart, then just ask.
fundapi-dsh-plugin
Command line
With a global dsh installed, drop the npx @deepseek-ai/dsh prefix.
npx @deepseek-ai/dsh plugin --profile web add fundapi-dsh-plugin
From GitHub
Same single command, without going through npm.
npx @deepseek-ai/dsh plugin --profile web add github:fundapi-cn/fundapi-dsh-plugin
Both the CLI and the Web UI Plugins page use the local pnpm (DSH does not bundle it; the desktop app ships its own) — run npm i -g pnpm if you don't have it. The desktop app manages its own profile and cannot be configured from the CLI; use the Plugins page there.
A key for the paid tools
fund_search, fund_profile andfund_screener_options work with no configuration at all. The other 16 (NAV history, holdings, position changes, managers, ratings, rankings, screening, look-through, dividends, trading rules) need a key:get one.
Option 1: environment variable
Add one line to ~/.dsh/.env (or under your $DSH_HOME):
FUND_KEY=cf-your-key
Option 2: profile config
Edit the profile's cordis.patch.yml:
- id: mcp-fundapi
name: fundapi-dsh-plugin
config:
apiKey: cf-your-key| Field | Default | Notes |
|---|---|---|
apiKey | reads FUND_KEY | Without it you run anonymously: 3 free tools only |
url | https://api.fundapi.cn/mcp | Point at a self-hosted proxy if you have one |
serverName | fundapi | Tool prefix; letters, digits, underscore and hyphen, 1–32 characters |
headers | empty | Extra request headers, merged with the key above |
Remaining fields (toolCallTimeoutMs, failOnStartupError, …) go to the MCP client bundled with DSH.
What you can ask
| You ask | Tool used |
|---|---|
| "Which fund code is "E Fund Consumer Sector"" | fund_search |
| "What are the fees, size and manager of 110022" | fund_profile |
| "How did this fund perform over the past year" | fund_nav |
| "What was added or trimmed last quarter" | fund_holding_changes |
| "Do my funds hold the same stocks" | fund_overlap |
| "What does my three-fund portfolio actually hold" | fund_portfolio_xray |
The full list of 19 tools is on the MCP setup page; field definitions live in the docs.
A few things that are easy to misread
- Before the daily NAV is published, what you get is an estimate, labelled as such — the published NAV is authoritative.
- Quarterly reports disclose only the top ten holdings, so that list is not the full portfolio, and "newly in the top ten" does not mean "newly bought".
- When an agency did not rate a fund in a period, the value is
null— that means not rated, not zero stars.
These are the tools' return conventions; how a model words the answer is up to the host. The judgement stays with you — the tools are there to get the data right.