Market Eyes Live MCP Server

Quick start

Claude Code

claude mcp add --transport http market-eyes-live https://marketeyeslive.com/mcp

Claude Desktop and VS Code

Add this to your MCP configuration file:

{
  "mcpServers": {
    "market-eyes-live": {
      "type": "http",
      "url": "https://marketeyeslive.com/mcp"
    }
  }
}

Cursor

Cursor infers a remote server from the presence of url, so it takes no type field:

{
  "mcpServers": {
    "market-eyes-live": {
      "url": "https://marketeyeslive.com/mcp"
    }
  }
}

Claude.ai and the Claude desktop app, as a connector

Settings, then Connectors, then add a custom connector and paste https://marketeyeslive.com/mcp. Leave the authentication fields empty.

ChatGPT

In ChatGPT, open Plugins, choose +, then Add custom MCP server, paste https://marketeyeslive.com/mcp and set authentication to none. Custom servers are available on the Plus, Pro, Business, Enterprise and Edu plans, and some accounts need developer mode turned on first. Menu names may vary as ChatGPT updates.

Perplexity

Settings, then Connectors, then add a custom connector with https://marketeyeslive.com/mcp as a remote server over Streamable HTTP, with no authentication. Custom connectors need a Perplexity Pro plan or above. Menu names may vary.

Gemini

In the Gemini web app at gemini.google.com, open the apps settings and add a custom app with https://marketeyeslive.com/mcp as the server address. Custom apps are offered on the Google AI Pro and Ultra plans. Menu names may vary.

Grok

On Grok on the web, go to grok.com/connectors, choose New Connector, then Custom, paste https://marketeyeslive.com/mcp and add no authentication. On Grok Business and Enterprise, a team admin provisions connectors first. Menu names may vary.

Mistral Le Chat

Open Intelligence, then Connectors, then add a Custom MCP Connector, paste https://marketeyeslive.com/mcp; Le Chat detects the authentication type, and this server needs none. Adding a connector may need a workspace admin. Menu names may vary.

Set it up on a computer; once added, it also works in the phone apps of Claude and ChatGPT.

That is the whole setup. There is no key to paste and no account to create.

Tools

The eight factor scores

Every rating shows its work through eight components: Valuation (price against fundamentals), Quality (balance sheet strength and profitability), Momentum (price trend and relative strength), Earnings (growth trajectory and surprises), Sentiment (analyst and market mood), Catalyst (near-term events that could move the stock), Risk-Adjusted (return for the risk taken), and Macro Fit (fit with the current market regime and sector cycle).

Questions it answers well

• "How does Market Eyes Live rate NVDA, and what does the rating rest on?"
• "Compare AAPL, MSFT and GOOGL on quality and valuation."
• "What are the flagged risks on TSLA today?"
• "How does VOO score on valuation right now?"
• "What is driving mortgage rates this week?"

What this is, and what it is not

Most financial MCP servers hand a model raw inputs: prices, fundamentals, news, technicals. The model then has to do the analysis itself, in-context, with no memory of whether its last read was any good.

This server hands back a finished analytical judgment: one conviction tier and one 0-100 composite, plus the eight factor scores that produced it. Two things back that number:

• The rules are stress-tested against 19 years of U.S. market history (2007 to 2026, spanning the 2008 financial crisis, the COVID crash, and the 2022 bear market), with delisted companies included so the test is not flattered by survivorship bias.
• Every published rating is forward-tested. Each one is recorded and graded daily against what the market actually did next, on a public record with a raw feed and a full validation study, including the ideas the testing program rejected.

To be explicit about the limits: Market Eyes Live does not issue buy or sell calls, this is research rather than financial advice, and the 19-year window tests the rules, not the live ratings, which are graded forward day by day. Coverage is U.S.-listed stocks and ETFs only.

How it compares to market-data MCP servers

These are complementary rather than competing. A raw data feed wins on breadth, history, and international coverage. This one wins when you want a scored read with a track record attached.

A fuller side-by-side, including when a raw data feed is the better choice: financial data MCP servers compared.

Usage is governed by the API terms of use: commercial use is fine, attribution is asked for, and nothing here is financial advice.

Coverage and rate limits

• Coverage: U.S.-listed stocks and ETFs only, more than 11,000 tickers. A core set is refreshed daily and answers instantly; anything else on the coverage list is scored live on demand. Crypto, futures, and non-U.S. listings (for example SHOP.TO or BTC-USD) return a "not covered" result. Five U.S. tickers that share their symbol with a futures contract or a coin (CORN, GOLD, WTI, BTC and ETH) are not rated, and the result says so. Class shares work in either spelling (BRK-B or BRK.B). A ticker that is not on the coverage list, such as a very recent listing, returns a result saying exactly that.
• What counts: only tool calls. Connecting, listing the tools and pings are free.
• Tool call limit: 240 calls per hour per user when the AI platform passes an anonymous user id (ChatGPT does), otherwise per network address. A platform that passes no user id, such as Claude, shares 3,000 calls per hour across all its users. Callers that do not come through ChatGPT or Claude share 3,000 calls per hour between them, so they can never use up the whole service. 6,000 calls per hour across all callers. Those are the limits in force as of 6 October 2026. They can change in either direction, and this page is the reference for what they currently are.
• Live on-demand scoring: 20 per hour per user or address, 100 per hour per platform, 150 per hour shared by callers that do not come through a platform, and 300 per hour across all callers, because those calls hit paid data vendors. Recent live scores are reused for 12 hours, and a ticker that could not be scored is not retried for 12 hours. Stating the shared caps because you can be limited by one while still under your own.
• When a limit is reached: the tool returns an error result that says which limit it was and when it resets, at the top of the hour (UTC).
• Asset-native scoring: an ETF is judged as a fund and a REIT on FFO, rather than forced through a common stock template.
• Limits are generous for normal agent use. If you are building something that needs more, email team@marketeyeslive.com.

One engine, two products

The MCP server and the Market Eyes Live app are the same company and the same engine. The MCP server is the free public data tier. The app carries the full research depth: suggested entry prices, alert levels, portfolio tools, a free paper-trading arena, and Lock Radar mortgage timing. It runs on iOS and the web.

Source, registry, and directories

• GitHub repository
• Official MCP registry: io.github.marketeyeslive/market-eyes-live
• Smithery and Glama
• llms.txt for AI crawlers

What is Market Eyes Live How MELANY is tested