Summary

A research desk you address in prose, with one escape hatch back to raw data.

The three agent tools take a prompt and hand back synthesis, which is both the point and the limit: what returns is Octagon's aggregation, not a dataset you can recompute or audit. prediction_markets_history is the exception, returning historical rows for a Kalshi event ticker with pagination and optional analysis columns. Four of the eight tools describe Octagon's own documentation rather than any market, so the research surface is narrower than the tool count first suggests.

What it is

A server over Octagon's hosted research agents, plus a live index of Octagon's own documentation; the research tools take a prompt and return synthesis rather than a dataset.

What you get

octagon-agent as the orchestrator across SEC filings, earnings calls, financial metrics, stock data, institutional holdings, private company research, funding, M&A, investor intelligence, debt and news
Multi-turn research: conversation and newConversation keep a thread going instead of restarting the context each question
octagon-deep-research-agent for questions that need aggregation across the web rather than one filing
Kalshi in two forms — an agent that reports what is driving prices and where market and model probabilities diverge, and prediction_markets_history for the raw series by event ticker, paginated with a capture-time window
Octagon's live documentation searchable, readable and refreshable in place, returned with source URLs

Requirements

OCTAGON_API_KEY. The research tools call Octagon's hosted agents, so every answer is a round trip billed to that account, while the docs tools read Octagon's public LLM-friendly endpoints. prediction_markets_history needs the Kalshi event ticker, which the tool describes as the final segment of the market URL.

Setup effort

One command plus a key — env OCTAGON_API_KEY=your_octagon_api_key npx -y octagon-mcp, then supply credentials