orcalayer-mcp

Model Context Protocol (MCP) server for the OrcaLayer API — Polymarket whale and market analytics inside Claude, Cursor and other MCP clients.

Two ways to use it:

• Hosted, nothing to install: https://orcalayer.com/mcp (Streamable HTTP). Add it as a connector in Claude (Settings → Connectors → Add custom connector, paste the URL) or in any client that supports remote MCP servers.
• Local (stdio): uvx orcalayer-mcp, for Claude Desktop and other local clients.

Either way it is a thin wrapper over the orcalayer Python SDK and exposes six read-only tools:

Public tools work anonymously. whale_alerts needs a Premium API key (get one): on the hosted server send it as the Authorization: Bearer <key> request header (clients that support request headers for a connector), on the local server set the ORCALAYER_API_KEY environment variable. Without a key the tool answers with a short notice, not an error.

Every tool carries readOnlyHint: true: nothing is created, changed or deleted on your behalf. Data is read from the OrcaLayer API only.

Each result carries a short notes block that defines its fields, so a model comparing two tools never meets two different numbers under one name: market_win_rate (share of resolved markets won) has the same basis in leaderboard and wallet_overview; profit_factor is capped at 99.99 on the leaderboard (profit_factor_capped) and uncapped net in the overview; indexed_volume_usd is the volume of fills in OrcaLayer's index, not Polymarket's figure; Smart Money counts are wallets, whatever 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 — live-derived settlement mechanics, shadow-verified 100.000% accurate on MINT/MERGE. It describes how the match settled, not trader intent (~80% of all fills settle as MINT; null = honest refusal, not "not a mint"). Details: the orcalayer://api-reference resource or orcalayer.com/docs/api.

Prompts

Ready-to-use prompts for common analytics scenarios:

In Claude Desktop, pick a prompt from the prompt menu (the + / slash-command picker) — each one orchestrates the tools above for you.

Resources

Read-only context the model can pull directly — no tool call needed:

Use with Claude Desktop

Add this to your claude_desktop_config.json (%APPDATA%\Claude\claude_desktop_config.json on Windows, ~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

{
  "mcpServers": {
    "orcalayer": {
      "command": "uvx",
      "args": ["orcalayer-mcp"]
    }
  }
}

The public tools work as-is. For the Premium whale_alerts tool, add your API key (get one):

{
  "mcpServers": {
    "orcalayer": {
      "command": "uvx",
      "args": ["orcalayer-mcp"],
      "env": { "ORCALAYER_API_KEY": "sk_orca_your_key_here" }
    }
  }
}

Restart Claude Desktop after editing the config.

Use the hosted server

The same six tools, prompts and resources are served at https://orcalayer.com/mcp over Streamable HTTP. No account is needed for the public tools. In Claude: Settings → Connectors → Add custom connector → paste the URL. For whale_alerts, send your Premium key as the Authorization: Bearer <key> request header where your client supports connector headers; keys are never accepted in the URL.

Self-host over HTTP

pip install orcalayer-mcp
orcalayer-mcp --http --host 127.0.0.1 --port 8020 --path /mcp

The server is stateless and answers plain JSON, so any reverse proxy with TLS in front of it 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 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 that carry a key always go to https://orcalayer.com. Put your own per-client rate limit in the proxy: the server itself does not limit callers.

License

MIT. See LICENSE.

Data is provided for informational purposes only and is not financial advice.

mcp-name: io.github.orcalayer/orcalayer-mcp