What is an MCP server for prediction markets?
A Model Context Protocol (MCP) server exposes a set of typed tools to AI agents — Claude Code, Cursor, Claude Desktop, and other MCP-compatible clients. For prediction markets that means tools like get_markets, get_orderbook, get_calibration, screen_indicators, create_thesis. The agent calls them by name; the MCP server hits the underlying venue (Kalshi, Polymarket, etc.) and returns normalized JSON. Without MCP, the agent would have to be taught each REST endpoint manually.
How do I install an MCP server in Claude Code?
For SimpleFunctions: run `claude mcp add simplefunctions --url https://simplefunctions.dev/api/mcp/mcp`. For npm-published servers: `npx -y <package-name>` and add the resulting command to your `claude_desktop_config.json` mcpServers map. For git-based servers: clone the repo, follow its README for the launch command, and reference that command in the same config. Restart your Claude client after editing the config.
Why does SimpleFunctions ship more tools than most single-venue MCP servers?
Most third-party MCPs cover one venue's read endpoints. SimpleFunctions covers Kalshi + Polymarket reads plus seven additional surfaces an agent typically wants on top of raw markets: a screener over computed indicators (implied yield, cliff risk, liquidity availability score), causal-tree thesis decomposition + signal injection, live calibration scores, world-state snapshots and deltas, news-stream subscriptions, a public agent-guide tool, and authenticated intent execution. The current count is published at /api/internal/statistics so the page does not rely on stale marketing copy.
Are these MCP servers free?
Most of the third-party servers on this list are MIT-licensed, no-auth, and free for personal use — the cost is just whatever rate limits the underlying venue (Polymarket public API, Kalshi public REST) imposes on unauthenticated traffic. SimpleFunctions reads through the public MCP endpoint are also free with no auth required; authenticated tools (thesis system, intents, autopilot) are free up to 15M LLM tokens, then pay-per-token. Baozi MCP and Clawmarket are closer to product subscriptions.
Which MCP server should I install if I want cross-venue arbitrage?
SimpleFunctions is currently the only MCP server that exposes a `cross_venue_pairs` tool returning matched contracts across Kalshi and Polymarket with live spreads. Other multi-venue servers (JamesANZ, prediction-markets-mcp) return separate venue feeds — you would need to write the matching logic yourself in the agent loop. For arbitrage workflows, SimpleFunctions's built-in pair table saves the canonical-question-normalization step.
Is Clawmarket the same as SimpleFunctions /openclaw?
No. Clawmarket (useclawmarket.com) is a third-party autonomous-trading agent product over Polymarket and Kalshi. SimpleFunctions /openclaw is a landing page for the Claude Skill packaging of SimpleFunctions's MCP tools — it is the SimpleFunctions product, not the Clawmarket service. The two are unrelated commercial offerings; we list Clawmarket here because it is sometimes confused with SimpleFunctions's open-source MCP surface.
How do I evaluate an MCP server before installing it?
Three checks: (1) read the source — is it doing what the README claims, or piping a third-party API through unaudited code? (2) check commit recency — MCPs depending on Polymarket / Kalshi public endpoints can break silently when those APIs change shape; an unmaintained server is a debugging cliff. (3) check the auth model — anything requiring wallet keys deserves a code audit before the first run.
Can I run multiple MCP servers at the same time?
Yes. Claude Code, Claude Desktop, and Cursor all support multiple MCP servers in parallel; each registers its tools under a namespaced prefix. A common pattern is SimpleFunctions for cross-venue normalized data plus a venue-specific MCP (e.g. guangxiangdebizi/PolyMarket-MCP) for tools that only that venue's native API exposes. Tool name collisions are resolved by the prefix.