Search "stock market MCP server" and most results are tutorials about building one yourself, or MCP servers that just wrap a generic price quote. Quantustik runs an actual finance MCP server in production — here is what's live, how it's discovered, and how an agent connects to it.
Model Context Protocol (MCP) is an open standard for connecting an AI agent to external tools and data over a structured interface, instead of the agent guessing how to scrape a web page or hand-roll an API integration for every service it needs. A server exposes a set of typed tools (function calls with defined inputs/outputs), resources, and guided prompts; any MCP-speaking client — Claude, or another compatible agent runtime — can discover and call them the same way regardless of which server it is talking to.
Quantustik hosts an MCP server at quantustik.com/mcp using the streamable-HTTP transport — no API key required to start, an optional free key raises rate limits. The server is discoverable through a standard manifest at /.well-known/mcp.json, which lists the transport, the anonymous rate limit, and where to get an optional key — exactly the metadata an agent needs to connect without a human reading docs first.
Every tool is callable anonymously under an hourly per-IP rate cap; an optional free API key raises the limits for heavier use.
More than a dozen tools are live today, covering swing signals, execution-ready trade plans, forecast probabilities across multiple horizons, market-conditions and fear/greed context, earnings-calendar risk, insider and institutional data, and a multi-ticker batch sweep so an agent can check a whole watchlist in one call instead of one round-trip per ticker. The tool set is intentionally described qualitatively here rather than with a hardcoded count, because it grows: call the server’s own `list_capabilities` tool to get the always-current list, or browse the manifest and developer docs.
A handful of guided prompts also ship on the server — pre-built workflows like a risk-first position check or a high-conviction swing-trade search — so an agent (or a human driving one) does not have to reinvent the right call order from scratch.
This is a read-only data-access layer, not a trading connection: no tool places or manages an order, and no autonomous trading is wired to it anywhere. It also does not make the underlying forecast more accurate — MCP is the pipe, not the signal. The same calibration caveats that apply to the rest of the product apply to every number an agent pulls through this server.
MCP is a data-access protocol, not a source of edge: it changes how an agent reaches Quantustik's data, not the underlying quantum forecast's own calibration limits. Every tool is read-only — no order is ever placed or managed through it. Every tool is callable without an API key under an hourly per-IP cap; an optional free key raises the limits. Current quotas are on the developers page, not restated here since they change.
Educational research only — not investment advice.