Reference

MCP Server

Alphonce speaks the Model Context Protocol over HTTP. Pick Claude (Anthropic), Gemini, ChatGPT, Grok, or Cursor below — the dropdown shows where to configure each client, what the connection looks like in that product, and the exact config to paste.

What is MCP?

The Model Context Protocol is an open standard for connecting AI assistants to data sources and tools. Alphonce's MCP server exposes your portfolio, research engine, and execution capabilities to any MCP-compatible client — using the model you already pay for.

1. Create a token

Go to Settings → Connected accounts and generate an Alphonce API token (aph_…). It is shown once. Tokens are read-only by default: portfolio, assets, market data and strategies. Nothing that places an order.

Included from Chui upwards

Connecting your own client is part of the Chui plan and above. Your model, your client, your key — the token only decides what it may read.

2. Choose your client

Same Alphonce server for every major AI client. What changes is where you paste the config and how tools show up in the UI.

Client

Where: Claude Desktop → Settings → Developer → Edit Config (claude_desktop_config.json). Or the Claude Code CLI from Anthropic.

What it looks like

After you restart, Alphonce appears in Claude’s tool picker. Ask about your portfolio and Claude calls Alphonce tools mid-reply — you’ll see tool names like GET_HOLDINGS in the run, then a normal answer in the chat.

Setup

1Create an Alphonce API token in Settings → Connected accounts.
2Open Claude Desktop → Settings → Developer → Edit Config.
3Paste the JSON below into claude_desktop_config.json (merge with any existing mcpServers).
4Restart Claude Desktop. Confirm Alphonce tools show in the tool picker.
5Optional — Claude Code: run the CLI one-liner in a terminal, then /mcp inside the session.

claude_desktop_config.json

{
  "mcpServers": {
    "alphonce": {
      "type": "streamableHttp",
      "url": "http://localhost:3000/api/mcp?apikey=aph_YOUR_TOKEN"
    }
  }
}

Claude Code CLI alternative:
claude mcp add --transport http alphonce "http://localhost:3000/api/mcp?apikey=aph_YOUR_TOKEN"

Try asking

What’s my portfolio concentration, and draft a rebalance plan toward my targets?

Endpoints

Remote MCP over HTTP — nothing to install locally unless a client requires a bridge.

Server	URL
Trading / portfolio	`http://localhost:3000/api/mcp`
Research	`http://localhost:3000/api/mcp/research`

Auth: Authorization: Bearer aph_…, or ?apikey=aph_… when the client cannot set headers (query strings can end up in logs — prefer the header when you can).

Discovery: /api/mcp/manifest.json. Full surface: API reference.

Available tools

Which tools appear depends on the scopes on your token. A read-only token will not expose PLACE_ORDER.

Tool	Description
`GET_PORTFOLIO_SUMMARY`	Total value, unrealised P&L, allocation, holding count
`GET_HOLDINGS`	Every position with cost basis and P&L
`SYNC_PORTFOLIO`	Pull the latest positions from a connected broker
`GET_TARGET_ALLOCATIONS`	Your target weights
`GENERATE_REBALANCE_PLAN`	Draft a plan to reach those weights
`GET_BOT_STATUS`	State of your bots
`GET_BOT_TRADES`	Trades a bot has made
`GET_BOT_ALERTS`	Alerts a bot has raised
`GET_COMPUTE_USAGE`	Compute Units consumed this period
`CREATE_STRATEGY`	Define a strategy (requires strategies:write)
`PLACE_ORDER`	Place an order (requires trades:execute — not on a read-only token)

Data privacy

Your token scopes what a client may reach, and scopes are checked on every call rather than only at connection time. Inference happens in your client with your model — Alphonce never sees the conversation. Revoke a token any time from Settings; it stops working immediately.