alpacahq/alpaca-mcp-server

> **Alpaca MCP Server v2 is here.** This version is a complete rewrite built with FastMCP and OpenAPI. If you're upgrading from v1, please read the Upgrade Guide — tool names, parameters, and configuration have changed.
Table of Contents
- Upgrading from V1
- Prerequisites
- Getting Your API Keys
- Setup
- Configuration
- Features
- Example Prompts
- Available Tools
- Testing
- Project Structure
- Troubleshooting
- Disclosure
---
Upgrading from V1
V2 is a **complete rewrite** built with FastMCP and OpenAPI. **None of the V1 tools exist in V2** — tool names, parameters, and schemas have changed. You cannot use V2 as a drop-in replacement if your setup depends on specific V1 tool names or parameters.
What changes
| Aspect | V1 | V2 |
| ------------------ | -------------------------------------- | -------------------------------------------------------------------------------------------- |
| **Tool names** | Hand-crafted (e.g. get_account_info) | Spec-derived with overrides (e.g. get_account_info — names may overlap but schemas differ) |
| **Parameters** | Custom schemas | Aligned with Alpaca API specs |
| **Configuration** | .env + init command | Env vars in MCP client config only |
| **Tool filtering** | Not supported | ALPACA_TOOLSETS env var |
| **Whitelisting** | Not supported | Use ALPACA_TOOLSETS to restrict tools |
How to avoid V1-style usage in V2
MCP clients discover tools dynamically from the server. There is no config file where you "whitelist" tool names — the client gets whatever tools the server exposes. To avoid your client or AI assistant using V2 incorrectly:
1. **Do not reuse V1 config** — Treat V2 as a new server. Update your MCP client config with the new command/args; remove any .env or init-based setup.
2. **Clear tool caches** — Restart your MCP client (Claude Desktop, Cursor, VS Code, etc.) after switching so it fetches the new tool list instead of using a stale one.
3. **Start a fresh chat/session** — Existing conversations may have cached references to old tool names. Start a new chat so the LLM sees the current V2 tools and their schemas.
4. **Update custom instructions and rules** — If you have Cursor rules, Claude instructions, or other prompts that mention specific V1 tool names (e.g. "use get_account_info"), update them to match V2 tool names or remove those references and let the LLM discover tools from context.
5. **Restrict tools with** ALPACA_TOOLSETS — If you previously limited which capabilities your assistant could use, V2 supports server-side filtering via the ALPACA_TOOLSETS env var. See Configuration > Toolset Filtering for the list of toolsets.
Summary
Assume **no backward compatibility** with V1. Reconfigure your MCP client for V2, restart it, and use a fresh session. Check the Available Tools section for the current tool list.
If you had custom V1 workflows
If you documented allowed tools, wrote scripts that call tools by name, or built prompts around specific V1 tool/parameter shapes — treat them as obsolete. Recreate them using the Available Tools listed below and the current parameter schemas exposed by the server.
Staying on V1
If you need to stay on V1, pin to the last V1 release (e.g. uvx alpaca-mcp-server==1.x.x serve) in your MCP client config. V1 remains available on PyPI for existing setups.
---
Prerequisites
- **Python 3.10+** (installation guide)
- **uv** (installation guide)
- **Alpaca Trading API keys** (free paper trading account)
- **MCP client** (Claude Desktop, Cursor, VS Code, etc.)
Getting Your API Keys
1. Visit the Alpaca Dashboard
2. Create a free paper trading account
3. Generate API keys from the dashboard
Setup
Add the server to your MCP client config, then restart the client. No init command, no .env files — credentials are set in **one place only**.
Claude Desktop
Edit ~/Library/Application Support/Claude/claude_desktop_config.json (Mac) or %APPDATA%\Claude\claude_desktop_config.json (Windows):
Claude Mobile
Claude Mobile requires a remote MCP server. This package does not configure remote MCP authentication, so do not expose it directly to the public internet with your Alpaca API keys. Use the local stdio setup from a supported desktop client.
ChatGPT
ChatGPT requires a remote MCP server. This package does not configure remote MCP authentication, so do not expose it directly to the public internet with your Alpaca API keys. Use the local stdio setup from a supported desktop client.
Cursor
Install from the Cursor Directory in a few clicks, or add to ~/.cursor/mcp.json: