Bringing Finance-Native Tools to Your AI Agents: The StockUp Quan MCP Server

General-purpose large language models are notoriously prone to hallucination when performing stock analysis. They make up financial ratios, miss subtle accounting red flags, and lack real-time market grounding. They approximate, compile, and conversationalize.

To solve this, we have launched the StockUp Quan MCP (Model Context Protocol) Server. This server connects local AI clients like Claude Desktop and Cursor directly to the StockUp Quan financial AI API. Instead of asking an agent to guess or write custom scrapers, the MCP server equips it with 10 finance-specific tools designed to produce grounded, auditable stock research.

Here is what the server introduces, how it changes the way you run research workflows, and what is next for remote platforms like ChatGPT.

Grounded Research and Adversarial Workflows

Rather than trying to do everything in a single prompt, the StockUp Quan MCP server structures analysis into separate, specialized tasks.

1. Realistic SEC Filing Audits

Using the analyze_sec_filing tool, your agent can review quarterly and annual reports for accounting signals, unrecognized liabilities, changes in management guidance, and earnings-quality issues. The tool extracts hard signals from the text, helping you identify red flags that standard summaries usually overlook.

2. Adversarial AI Research

Grounded data is only half the battle; the reasoning itself must be challenged. The get_quan_research_debate_packet tool enables a two-agent review process. When called, it does not just return a polished conclusion. Instead, it generates a structured debate packet containing:

• Conditional Claim: An evidence-first investment thesis with explicit confidence or uncertainty levels.
• Verified Evidence & Assumptions: The underlying data and assumptions that support the claim.
• Counter-Case & Falsifiers: The strongest alternative interpretation and specific triggers that would prove the main thesis wrong.
• Review Questions: 3 to 6 targeted questions for the host agent.

This allows your local agent (such as Claude or Cursor) to act as a critical reviewer—actively questioning the assumptions and analyzing contrary evidence before you make a decision.

Installation

You can run the server locally using the stdio transport.

1. Generate your API key at stockup.cc/api.
2. Add the configuration to your MCP client settings:

{
  "mcpServers": {
    "stockup-quan": {
      "command": "npx",
      "args": ["-y", "@stockup/quan-mcp-server"],
      "env": {
        "STOCKUP_API_KEY": "sk_quan_your_key"
      }
    }
  }
}

3. Restart your client. If you prefer to run or inspect the source directly, the repository is open at TrecoOffical/stockup-mcp.

Coming to the ChatGPT GPT Store

The current release runs locally via stdio commands. Because of this, it is built for desktop clients like Claude Desktop or Cursor that run code directly on your machine.

ChatGPT operates entirely in the cloud and cannot run local command-line packages. To bring these capabilities to ChatGPT users, we are developing a remote version of the StockUp Quan MCP server.

This remote version will connect to ChatGPT via webhooks and will use StockUp OAuth. This allows you to securely sign in and link your StockUp account to custom GPTs in the GPT Store without pasting raw API keys into chat prompts. We expect to launch this integration soon.