Plug Orrery into your agent

Use the local package when a client expects a command-based MCP server instead of a remote URL. The wrapper speaks standard Content-Length framed stdio MCP locally, then forwards each JSON-RPC call to Orrery over HTTPS.

{
  "mcpServers": {
    "orrery": {
      "command": "npx",
      "args": ["-y", "@orrery-labs/mcp"],
      "env": {
        "ORRERY_API_KEY": "orrery_live_..."
      }
    }
  }
}

API keys are forwarded as X-Orrery-API-Key. Paid x402 proofs can also be forwarded with ORRERY_X_PAYMENTand ORRERY_PAYMENT_SIGNATURE.

Open Settings → Developer, click Edit Config, and add:

{
  "mcpServers": {
    "orrery": {
      "url": "https://orrery.me/api/mcp/v1"
    }
  }
}

Restart Claude Desktop. You'll see Orrery tools, resources, and prompts appear in the picker — orrery_brief_today,orrery_market_why,orrery_search, and so on.

These are the first workflows to wire into Claude, Cursor, newsroom assistants, or internal research agents. Each one keeps Orrery in verification mode, not trade advice mode.

Edit ~/.cursor/mcp.json (create it if it doesn't exist):

{
  "mcpServers": {
    "orrery": {
      "url": "https://orrery.me/api/mcp/v1"
    }
  }
}

Reload Cursor — the agent panel surfaces Orrery tools alongside your local stdio servers.

The Anthropic SDK supports MCP transports natively. Pass the URL when constructing your agent and Claude will pick the tools up automatically.

from anthropic import Anthropic
from anthropic.lib.mcp import HTTPMCPClient

mcp = HTTPMCPClient(url="https://orrery.me/api/mcp/v1")
client = Anthropic()

resp = client.messages.create(
    model="claude-opus-4-7",
    max_tokens=600,
    thinking={"type": "adaptive"},
    mcp_servers=[mcp],
    messages=[{
        "role": "user",
        "content": "Summarise today's biggest prediction-market move "
                   "and tell me what to verify before trusting it."
    }],
)
print(resp.content[0].text)

Each tool maps 1:1 onto an x402 endpoint. Names are namespaced under orrery_ so they sort cleanly in the tool picker. Pricing matches the catalog and paid calls require x402 settlement.

• orrery_decision_attention$0.05
• orrery_decision_market$0.15
• orrery_brief_today$0.01
• orrery_markets_movers$0.005
• orrery_venuesfree
• orrery_markets$0.005
• orrery_divergence_top$0.05
• orrery_market_snapshot$0.005
• orrery_market_why$0.02
• orrery_market_resolution_risk$0.01
• orrery_event_cluster$0.03
• orrery_signals$0.01
• orrery_signals_kind$0.01
• orrery_search$0.005
• orrery_events_top$0.005
• orrery_share_card$0.03
• orrery_wallet$0.02
• orrery_category_intelligence$0.02
• orrery_backtest$0.02
• orrery_research_anomalies$0.05
• orrery_trades_recent$0.005
• orrery_watchlist_summary$0.05
• orrery_portfolio_risk$0.05
• orrery_healthfree

Resources give agents stable Orrery context without spending a paid tool call. Prompts provide safe workflow scaffolds for daily research, market due diligence, resolution checks, and cross-venue comparison.

# 1. Initialize handshake
curl -s https://orrery.me/api/mcp/v1 \
  -X POST -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize",
       "params":{"protocolVersion":"2024-11-05",
                 "capabilities":{},"clientInfo":{"name":"curl","version":"1"}}}'

# 2. List tools
curl -s https://orrery.me/api/mcp/v1 \
  -X POST -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/list"}'

# 3. Call a tool
curl -s https://orrery.me/api/mcp/v1 \
  -X POST -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":3,"method":"tools/call",
       "params":{"name":"orrery_brief_today","arguments":{}}}'

# 4. List resources and prompts
curl -s https://orrery.me/api/mcp/v1 \
  -X POST -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":4,"method":"resources/list"}'

curl -s https://orrery.me/api/mcp/v1 \
  -X POST -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":5,"method":"prompts/list"}'

Orrery's MCP initialize handshake instructs your agent to refuse any request to turn these tools into directional trade instructions. The tools surface public Polymarket data with explicit calibration tiers; never predicted outcomes, never insider feeds, never copy-trading advice. The same refusal language is in/llms.txt and the agent plugin manifest.

Next steps

• • CLI version of the same surface: /docs/agents/cli
• • Raw HTTP without MCP: /docs/agents/curl
• • Service catalog (Agentic Market discovery): /.well-known/x402-services.json