Skip to main content
Use this page after Client setup if you want an optional guided workflow for Centaur in agent-capable clients.

Overview

The public Centaur skill gives agent clients a guided way to use Centaur over MCP after the Centaur server is already configured. It is a simple workflow layer on top of the connected Centaur server, not a replacement for the underlying REST and MCP reference.

Before you begin

Before you install the skill, make sure this is already true:
  • Centaur is configured in your client through MCP
The skill does not run the OAuth flow or dynamic registration itself. It assumes your client has already connected to the Centaur MCP server.

Install

Canonical repo:

How the skill works

The skill follows a standard decision path:
  1. Prefer MCP if Centaur is already configured in the current client.
  2. If MCP is not configured, stop and give setup instructions.

Supported surfaces

  • MCP endpoint: https://partners.centaur.io/mcp
  • MCP access to Centaur read families across the partner feed, events, messages, generated summaries, positions, discovery, stats, rankings, and activity summaries

Common reads

  • partner feed
  • events
  • messages
  • generated narrative summaries
  • positions
  • discovery
  • stats
  • trader rankings
  • activity summaries
Generated narrative summaries are for market narrative questions. They should not be used as evidence for exact trade counts, public activity rankings, or current open-position skew; use rank_traders for activity and performance rankings, summarize_message_activity for message/event counts and trends, list_traders.tradeCount for trader discovery activity counts, and event, message, position, open-position, and stat tools for other trade facts and positioning claims.

Authentication

The skill uses the client-managed OAuth connection to the Centaur MCP server. It does not create API keys or run the OAuth flow itself.

Notes

  • The connected Centaur MCP server remains the source of truth for the exact live tool inventory and resource URIs.
  • Set up the MCP server first, then install the skill for guided prompts and workflow behavior.

Supported clients

  • Claude
  • ChatGPT
  • Cursor
  • Codex
  • Claude Code
Set up Centaur in one of those clients first, then install the skill so the agent can use the connected MCP server.

Repository contents

  • root SKILL.md
  • references/auth.md
  • references/mcp.md
  • references/rest.md
  • references/examples-curl.md
  • references/client-setup.md

Next steps

Resources

  • Public skill repo:
  • Machine discovery:
  • Install command:
Last modified on August 6, 2026