Portfolio Research MCP Server 📈

A Model Context Protocol (MCP) server that gives any AI coding assistant or chat client live stock and portfolio research tools, powered by Yahoo Finance data. Written in Python with FastMCP, it speaks Streamable HTTP so you can host it once and connect from Claude, Cursor, VS Code, Google Antigravity, Windsurf, and terminal-based agents.

Ask your assistant: "Analyze my portfolio: 10 AAPL bought at $150 and 5 MSFT at $300," and it will call this server, price the holdings live, and report your gain/loss and weights.

⏳ First request is slow — please read

The public demo is hosted on Render's free tier, which puts the server to sleep after ~15 minutes of inactivity. When you send your first request after it has been idle, Render has to wake it up (a "cold start"), which takes roughly 30–60 seconds.

• Your first message may hang or time out — this is normal. Just wait and try again; once awake, the server responds instantly.

• Tip: open https://stock-research-mcp.onrender.com/health in your browser first. When it shows ok, the server is warm and ready.

• This delay only affects the free hosted demo. A local run or a paid Render instance has no cold start.

Related MCP server: Finance MCP Server

🚀 Try it instantly (public demo)

A live instance is already hosted — you don't need to deploy anything to test it.

Drop those into any config block in Connect it to your IDE / agent below and start asking questions.

⚠️ This is a shared public demo token for trying the server out. It may be rate-limited, rotated, or taken down at any time, and the instance sleeps after ~15 min idle (first call may take 30–60s to wake). For anything real, deploy your own and set your own private MCP_AUTH_TOKEN.

Tools

Data source: yfinance — free, no API key required.

Quick start (local)

Requires Python 3.10+.

git clone <your-repo-url> stock-research-mcp
cd stock-research-mcp

python -m venv .venv
# Windows:
.venv\Scripts\activate
# macOS/Linux:
source .venv/bin/activate

pip install -r requirements.txt
python server.py

The server starts at:

• MCP endpoint: http://127.0.0.1:8000/mcp

• Health check: http://127.0.0.1:8000/health → ok

Locally, auth is disabled unless you set MCP_AUTH_TOKEN (see Security).

Deploy free on Render

Render's free web-service tier hosts this at a public HTTPS URL. A render.yaml blueprint is included.

1. Push this repo to GitHub.

2. In the Render dashboard: New + → Blueprint, select your repo. Render reads render.yaml and provisions a free web service. (Or New + → Web Service manually: Build pip install -r requirements.txt, Start python server.py.)

3. Under Environment, set a secret MCP_AUTH_TOKEN (a long random string).

4. Deploy. Your server is live at:

   https://<your-app-name>.onrender.com/mcp

⚠️ Free-tier cold starts: free instances spin down after ~15 minutes idle. The next request wakes it and can take 30–60 seconds. Upgrade to any paid instance to stay always-on. Hit /health to warm it before a demo.

Connect it to your IDE / agent

The blocks below use the public demo URL and token so you can copy-paste and go. To use your own deployment, swap in your Render URL and your private MCP_AUTH_TOKEN (or http://127.0.0.1:8000/mcp with no header for a local run).

⚠️ The config key differs by client. Cursor and VS Code use url; Antigravity and Windsurf use serverUrl; Claude Code uses a CLI command. Copy the right block below.

Claude Code (terminal)

claude mcp add --transport http stock-research \
  https://stock-research-mcp.onrender.com/mcp \
  --header "Authorization: Bearer mYfUEnx-jnDKlUTHIkUXhCWhMBPgxuzmSvCfxegrt98"

Then check it: claude mcp list. Remove with claude mcp remove stock-research.

Claude Desktop

Claude Desktop connects to remote servers through the mcp-remote bridge (needs Node.js). Edit claude_desktop_config.json (Settings → Developer → Edit Config):

{
  "mcpServers": {
    "stock-research": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote",
        "https://stock-research-mcp.onrender.com/mcp",
        "--header", "Authorization: Bearer mYfUEnx-jnDKlUTHIkUXhCWhMBPgxuzmSvCfxegrt98"
      ]
    }
  }
}

Restart Claude Desktop. Tools appear under the 🔌 (MCP) icon.

Cursor

Edit ~/.cursor/mcp.json (global) or .cursor/mcp.json (per-project):

{
  "mcpServers": {
    "stock-research": {
      "url": "https://stock-research-mcp.onrender.com/mcp",
      "headers": {
        "Authorization": "Bearer mYfUEnx-jnDKlUTHIkUXhCWhMBPgxuzmSvCfxegrt98"
      }
    }
  }
}

Then Settings → MCP and confirm the server shows green.

Google Antigravity

Antigravity uses serverUrl (not url). Open the MCP store via the "…" menu in the agent panel → Manage MCP Servers → View raw config, or edit ~/.gemini/config/mcp_config.json directly:

{
  "mcpServers": {
    "stock-research": {
      "serverUrl": "https://stock-research-mcp.onrender.com/mcp",
      "headers": {
        "Authorization": "Bearer mYfUEnx-jnDKlUTHIkUXhCWhMBPgxuzmSvCfxegrt98"
      }
    }
  }
}

Save, then Customizations → Installed MCP Servers → Refresh.

Windsurf

Windsurf also uses serverUrl. Edit ~/.codeium/windsurf/mcp_config.json (or Settings → Cascade → MCP Servers → Manage → View raw config):

{
  "mcpServers": {
    "stock-research": {
      "serverUrl": "https://stock-research-mcp.onrender.com/mcp",
      "headers": {
        "Authorization": "Bearer mYfUEnx-jnDKlUTHIkUXhCWhMBPgxuzmSvCfxegrt98"
      }
    }
  }
}

Click Refresh in the MCP panel.

VS Code (Copilot agent mode)

Create .vscode/mcp.json in your workspace (or run MCP: Add Server from the Command Palette):

{
  "servers": {
    "stock-research": {
      "type": "http",
      "url": "https://stock-research-mcp.onrender.com/mcp",
      "headers": {
        "Authorization": "Bearer mYfUEnx-jnDKlUTHIkUXhCWhMBPgxuzmSvCfxegrt98"
      }
    }
  }
}

Open Copilot Chat, switch to Agent mode, and the tools become available.

Other terminal / stdio-only agents

Clients that speak the remote MCP shape can reuse the Cursor block above (url + headers). For agents that only support local stdio servers, bridge to the remote URL with mcp-remote:

npx -y mcp-remote https://stock-research-mcp.onrender.com/mcp \
  --header "Authorization: Bearer mYfUEnx-jnDKlUTHIkUXhCWhMBPgxuzmSvCfxegrt98"

Point the client's command/args at that (as in the Claude Desktop example).

Usage examples

Once connected, try prompts like:

• "What's AAPL trading at right now?" → get_quote

• "Compare NVDA, AMD and INTC on valuation." → compare_tickers

• "Show me TSLA's price history over the last 6 months." → get_price_history

• "Any recent news on MSFT?" → get_news

• "Analyze my portfolio: 10 AAPL at $150, 5 MSFT at $300, 20 NVDA at $110." → analyze_portfolio

Security

• If MCP_AUTH_TOKEN is set, every request to /mcp must send Authorization: Bearer <token> or it's rejected with 401. Always set it in production.

• If it's unset, auth is off — fine for localhost, never for a public URL.

• A shared bearer token is a pragmatic choice for a personal/portfolio project. The MCP spec's full answer for public servers is OAuth 2.1 with PKCE; adopt that if you expose this to untrusted clients.

• The token shown in this README is a deliberately public demo token so anyone can try the hosted instance. Never reuse it for a private deployment — generate your own with python -c "import secrets; print(secrets.token_urlsafe(32))" and keep it secret.

Project structure

.
├── server.py          # FastMCP app: tools, bearer-auth middleware, /health, HTTP transport
├── portfolio.py       # yfinance data layer (pure, testable, returns plain dicts)
├── requirements.txt   # fastmcp, yfinance, uvicorn
├── render.yaml        # Render blueprint (free web service)
├── .env.example       # MCP_AUTH_TOKEN, PORT
└── README.md

Troubleshooting

License

MIT — do whatever you like. Data is provided by Yahoo Finance via yfinance and is subject to their terms; this project is for research/educational use, not investment advice.