How do I connect an agent to the data?
Two ways. Add the MCP server URL to any MCP-compatible client and authenticate with your API key, or call the REST API directly and generate a client from the OpenAPI 3.1 spec.
Your agent is only as good as its tools. RyxelData ships an MCP server and a JSON REST API designed for LLM tool-use from the first request.
Today
The agent scrapes EDGAR HTML, guesses at PDFs, or answers from training data. Every parsing miss becomes a confident wrong number.
With RyxelData
Typed MCP tools and a JSON REST API, scoped to the datasets on your account. Answers cite the Form 4, the 13F, or the S-1.
An agent that answers questions about insider trading or institutional ownership is only as reliable as its data source. Today that source is usually a search engine, a PDF, or an EDGAR page built for a human with a browser.
Scraping works until it doesn't. EDGAR's HTML shifts, PDF tables break parsers, and every edge case your pipeline mishandles becomes a plausible-sounding wrong answer. In finance, a hallucinated number is not a bug users forgive.
The alternative is wiring each source yourself: auth, schemas, retries, rate limits, one integration per dataset. That is months of plumbing before the agent does anything interesting.
MCP
Add one server URL. The client gets typed tools for insider trades, 13F positions, Form 144 notices, fund holdings, supply chains, and IPOs.
REST
On a custom stack, generate a client from the public OpenAPI 3.1 spec. The agent's contract with the data is explicit.
FORM 4
The tool most agents hit first. Each record includes the EDGAR link, so the answer can cite the filing.
SCOPE
The MCP server exposes tools for the subscriptions on the account. A key scoped to insider data cannot touch the rest.
Connect the RyxelData MCP server and your agent gets typed tools for insider transactions, 13F holdings, Form 144 proposed sales, fund holdings, supply chain relationships, and IPO data. Add one URL to Claude, Cursor, Claude Code, or OpenCode and the tools appear.
On a custom stack, the same datasets sit behind a REST API with a public OpenAPI 3.1 spec. Your framework can generate a client and validate responses against the schema, so the agent's contract with the data is explicit.
Responses are normalized JSON with stable field names, and every record links back to the source filing on EDGAR. The agent reads filing data like any other tool output, and it can cite the primary source.
Answers come from real filings instead of parametric memory. When a user asks who bought shares last week, the agent queries the Form 4 record and cites it.
Access stays scoped. The MCP server exposes only the datasets on your account, and the API key lives in the OAuth flow, not in the chat context.
You ship the agent. We keep the parsers running.
If your product is the agent, the data layer should be a tool you connect, not a second company you run. RyxelData sits next to your model, your orchestration, and your evals.
Claude, Cursor, Claude Code, and OpenCode. One URL, OAuth with your API key, tools appear in the client.
LangGraph, custom tool routers, your own runtime. Same JSON, same OpenAPI spec, same source links.
The key lives in the OAuth flow or your backend, never in the prompt. Access follows the account, not the conversation.
Two ways. Add the MCP server URL to any MCP-compatible client and authenticate with your API key, or call the REST API directly and generate a client from the OpenAPI 3.1 spec.
Claude, Cursor, Claude Code, and OpenCode are covered in the docs. Any client that speaks the Model Context Protocol can connect.
Through an OAuth flow tied to your RyxelData API key. The key never enters the chat context or the model's prompt.
Yes. The MCP server exposes tools only for the datasets on your account, so access stays scoped to what you subscribe to.
Share your use case and the datasets you need — we'll set up access for your team and one API key that covers it all.