PredictAsiaX MCP for Claude — integration guide

The Model Context Protocol (MCP) is an open standard released by Anthropic in November 2024 that lets AI assistants (Claude Desktop, Claude Code, and any MCP-compatible client) connect to external data sources and tools. MCP servers expose tools, resources, and prompts over stdio or HTTP transport. Once configured, the AI can invoke the server's tools directly during conversation.

Once connected, Claude can: (1) search live prediction markets and read market detail, order books, price history, and category/volume leaderboards; (2) fetch top probability movers over configurable time windows; (3) read a user's portfolio, positions, orders, and balances (with API key); (4) place, cancel, and batch-cancel orders (with API key + explicit user confirmation); (5) retrieve resolution evidence for settled markets; (6) estimate fees, manage subaccounts, post market-maker quotes and RFQs, submit market proposals, and manage builder apps, webhooks, and access requests (with API key); (7) run sandbox agent example flows for development. All 44 tools are documented at mcp.predictasiax.com/mcp/manifest.json.

Add the following to your Claude Desktop config at ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows). Include an 'mcpServers' object with a pax entry pointing to https://mcp.predictasiax.com/mcp/sse for SSE transport, or use the HTTP endpoint /mcp/call for stateless invocation. Restart Claude Desktop and the pax tools appear in the tools list.

17 of the 44 tools are public and require no authentication — these cover market data and reference lookups (market_search, market_orderbook, probability_movers, fee_estimate, resolution_evidence, sandbox_agent_examples, and others). The other 27 touch a specific user's account — portfolio, orders, subaccounts, market-making, builder apps, and webhooks — and require an API key issued from the PredictAsiaX Builders portal at builders.predictasiax.com. API keys are passed as Authorization: Bearer sk_live_* on every request. See docs.predictasiax.com/auth for the full auth guide.

Yes. Claude Code (the CLI) supports the same claude_desktop_config.json format via ~/.config/claude/mcp-config.json (or the CLI's --mcp-config flag). Once configured the MCP tools are available via the Skill and Bash tool wrappers in interactive and non-interactive Claude Code sessions.

PredictAsiaX provides official Python and TypeScript SDKs that wrap the REST + WebSocket API with HMAC signing built in. Python: pip install predictasiax; TypeScript: npm install @predictasiax/sdk. Both are open-source and follow industry-standard patterns developers expect from prediction market SDKs. Source and examples at github.com/predictasiax (currently under appeal for account restoration).

The REST API (docs.predictasiax.com/api) is a general HTTP interface for any client. The MCP server is a thin adapter on top of the REST API optimized for AI agents — it exposes the same operations as MCP tools with descriptions and JSON schemas that AI models can reason about, and it handles auth, rate limits, and idempotency headers automatically. For AI-driven workflows, use MCP; for traditional backend integration, use REST directly.

Yes. Public tools: 60 requests per minute per IP. Authenticated tools: 300 requests per minute per API key, higher tiers available for Builder Program participants. When rate-limited the response returns 429 with a Retry-After header. See docs.predictasiax.com/security for full rate limit policy.