MCP Server

What is the MCP server?

The Quant Data MCP server lets Claude, Cursor, and any other Model Context Protocol client query our data in plain English. Paste one URL and your existing API key into your client's config, and every REST endpoint becomes a tool the model can call on your behalf.

The Model Context Protocol is an open standard that lets large language models invoke tools the same way you would call a REST endpoint. When you ask the model a market question, it reads the catalog of available tools, picks the right one, fills in its parameters, and shows you the answer alongside the structured data it pulled.

Our server is hosted, so there is nothing to install, nothing to run locally, and no separate credential to manage. Your Quant Data API key is the only thing your client needs. If you have never used our REST API, the MCP is often the fastest way to start. If you already use REST, the MCP gives you a conversational front-door to the exact same data without writing more glue code.

Why connect it

Four representative workflows that work well as conversations with a model holding all of our tools.

Server details

Where to point your client, how authentication works, and how quota is shared with the REST API.

Endpoint

Streamable HTTP, hosted on the same domain as the REST API. Point any MCP-capable client at this URL and you are connected.

Authentication

Each request carries your Quant Data key in the Authorization header using the Bearer scheme. The MCP does not store keys, does not maintain sessions, and does not issue tokens of its own. Generate or rotate keys from the API keys page in your dashboard.

Quota and billing

MCP tool calls draw from the same per-user request quota as direct REST calls. One tool call equals one request against your bucket. You can mix MCP and REST traffic on the same key without splitting subscriptions or paying twice.

Errors

When something is off (a missing key, an unknown ticker, a malformed filter, a tripped rate limit) the server surfaces the same RFC 9457 problem detail you would see on the REST side, with the field names and remediation hints intact so the model can correct itself on the next turn.

Connect your client

Pick your client below and open it for the exact config. Any other client that speaks streamable HTTP works the same way: point it at the URL, set the Authorization header, and restart the client so the tool list refreshes.

{
  "mcpServers": {
    "quantdata": {
      "type": "http",
      "url": "https://api.quantdata.us/mcp",
      "headers": {
        "Authorization": "Bearer <YOUR_API_KEY>"
      }
    }
  }
}

claude mcp add --transport http -s user quantdata https://api.quantdata.us/mcp \
  --header "Authorization: Bearer <YOUR_API_KEY>"

Name              Quant Data
MCP Server URL    https://api.quantdata.us/mcp
Authentication    Custom header   ← only if the form exposes one
Header name       Authorization
Header value      Bearer <YOUR_API_KEY>

{
  "mcpServers": {
    "quantdata": {
      "url": "https://api.quantdata.us/mcp",
      "headers": {
        "Authorization": "Bearer <YOUR_API_KEY>"
      }
    }
  }
}

{
  "servers": {
    "quantdata": {
      "type": "http",
      "url": "https://api.quantdata.us/mcp",
      "headers": {
        "Authorization": "Bearer <YOUR_API_KEY>"
      }
    }
  }
}

{
  "mcpServers": {
    "quantdata": {
      "serverUrl": "https://api.quantdata.us/mcp",
      "headers": {
        "Authorization": "Bearer <YOUR_API_KEY>"
      }
    }
  }
}

{
  "mcpServers": {
    "quantdata": {
      "type": "streamableHttp",
      "url": "https://api.quantdata.us/mcp",
      "headers": {
        "Authorization": "Bearer <YOUR_API_KEY>"
      }
    }
  }
}

{
  "mcpServers": {
    "quantdata": {
      "httpUrl": "https://api.quantdata.us/mcp",
      "headers": {
        "Authorization": "Bearer <YOUR_API_KEY>"
      }
    }
  }
}

extensions:
  quantdata:
    enabled: true
    type: streamable_http
    name: quantdata
    uri: https://api.quantdata.us/mcp
    timeout: 300
    headers:
      Authorization: Bearer <YOUR_API_KEY>

{
  "transport": "streamable-http",
  "url": "https://api.quantdata.us/mcp",
  "headers": {
    "Authorization": "Bearer <YOUR_API_KEY>"
  }
}

The tool catalog

Twenty REST endpoints are mirrored as MCP tools, named qd_get_<concept> in snake_case. The twenty-one tools below cover the order-flow, exposure, statistics, volatility, and price-history surfaces, plus a small health-check helper. The newer REST-only endpoints (open-interest, max-pain, dark-pool levels, exchange notifications, market map, news articles) are reachable through the REST API today and will pick up MCP wrappers in a future release. The model sees a detailed description for each tool below, so you almost never need to name one by hand: describe what you want and the model picks for you.

Each tool exposes the same parameters as its REST counterpart: convenience filters such as tickers, strikePrices, and expirationDates, the filter expression DSL for boolean composition, time selection through sessionDate or a timeRange, pagination via size and searchAfter on the tape tools, and field projection with includes and excludes.

How a tool call flows

The MCP and the REST API are the same surface reached two different ways.

Because the MCP is stateless, each tool call is independent. Your client may run several in parallel, and none of them share session state on our side. If your conversation has the model fan out across twelve ticker queries at once, you will see twelve quota draws and twelve usage events.

Asking good questions

The model picks tools by reading their descriptions and matching them to the way you phrased the question. A few habits make that match noticeably better.

None of this is required: the model will still produce useful answers on vague prompts. Specific phrasings, named time windows, and a sense for which tool family fits the question all reduce wasted tool calls.

Where to go next