Purpose
OrcaLayer MCP is a Model Context Protocol (MCP) server used to connect the OrcaLayer API to AI agents, thereby providing Polymarket whale tracking and market analysis data in Claude, Cursor, Copilot, and other MCP clients. It is essentially a lightweight wrapper around the orcalayer Python SDK.
Main Capabilities
The server provides six read-only tools, of which the public tools can be used anonymously. The whale_alerts tool requires a Premium API key. Each tool carries readOnlyHint: true, meaning that nothing is created, changed, or deleted on behalf of the user; data is only read from the OrcaLayer API. Each result includes a short notes block that defines its fields, thereby ensuring that the model does not encounter two different numbers under the same name when comparing two tools. For example: market_win_rate (the proportion of wins in settled markets) has the same basis in leaderboard and wallet_overview; profit_factor is capped at 99.99 in leaderboard (profit_factor_capped), while in overview it is the uncapped net value; indexed_volume_usd is the turnover in the OrcaLayer index, not Polymarket's figure; Smart Money count is the number of wallets, regardless of their size. New (2026-07-22): The Premium SSE stream (/api/public/v1/live/trades) now carries settlement_type (MINT / MERGE / COMPLEMENTARY / null) on every event — the settlement mechanism derived in real time, shadow-verified to reach 100.000% accuracy on MINT/MERGE. It describes how a match settles, not trader intent (about 80% of all fills settle as MINT; null = an honest refusal, not "not a mint"). For details, see the orcalayer://api-reference resource or orcalayer.com/docs/api. In addition, ready-to-use prompts for common analysis scenarios are also provided: in Claude Desktop, select a prompt from the prompt menu (+ / slash command selector) — each prompt orchestrates the above tools for you. The model can directly pull read-only context without tool calls.
How to Use
There are two ways to use it:
- Hosted service, no installation required:https://orcalayer.com/mcp（Streamable HTTP). In Claude, add it as a connector (Settings → Connectors → Add custom connector, paste the URL), or add it in any client that supports remote MCP servers.
- Local (stdio): uvx orcalayer-mcp, suitable for Claude Desktop and other local clients.
Use with Claude Desktop
Add the following to your claude_desktop_config.json (%APPDATA%\Claude\claude_desktop_config.json on Windows, ~/Library/Application Support/Claude/claude_desktop_config.json on macOS). The public tools can be used directly. For the Premium whale_alerts tool, add your API key (get one). Restart Claude Desktop after editing the configuration.
Use the hosted server
The same six tools, prompts, and resources are available via Streamable HTTP at https://orcalayer.com/mcp Available at. Public tools require no account. In Claude: Settings → Connectors → Add custom connector → paste the URL. For whale_alerts, where the client supports connector request headers, send your Premium key as the Authorization: Bearer <key> header; keys are never accepted in the URL.
Self-hosting over HTTP
The server is stateless and returns plain JSON, so any reverse proxy with TLS will do (forward POST /mcp to 127.0.0.1:8020/mcp). The ORCALAYER_MCP_HTTP=1, ORCALAYER_MCP_HOST, ORCALAYER_MCP_PORT, and ORCALAYER_MCP_PATH environment variables are equivalent to the corresponding flags. ORCALAYER_MCP_ANON_BASE_URL optionally routes anonymous calls to a different API host (the hosted server points it at its local backend); calls carrying a key are always sent to https://orcalayer.com。请在代理中设置自己的每客户端速率限制：服务器本身不限制调用者。
When to use
Suitable for users who need Polymarket whale tracking and smart money data in AI agents, for example for analysis in MCP clients such as Claude, Cursor, and Copilot.
Limitations
This project is under the MIT license, see LICENSE for details. The data is for reference only and does not constitute financial advice. The whale_alerts tool requires a Premium API key; without a key, the tool returns a brief notice instead of an error.
