Kalshi MCP Server
DepthFeed runs a remote Model Context Protocol (MCP) server, so Claude, Cursor, Codex, and ChatGPT can pull Kalshi order-book depth and price data mid-session — no SDK, no glue code. It is keyless to start: add the endpoint and your agent can query BTC Kalshi markets immediately, then drop in a key for everything else.
The DepthFeed MCP server is a remote Streamable-HTTP endpoint (https://api.depthfeed.com/mcp) that exposes Kalshi order-book data as Model Context Protocol tools. AI agents call tools like polymarket_search_markets and kalshi_get_orderbook to fetch live and historical depth as JSON, with the same plan gating as the REST API.
Kalshi MCP server at a glance
- Endpoint
- https://api.depthfeed.com/mcp
- Transport
- Remote MCP — Streamable HTTP
- Auth
- Keyless demo, or Bearer df_ key
- Tools
- 11 read-only — search · order book · history · screener
- Assets
- 7 — BTC · ETH · SOL · XRP · DOGE · BNB · HYPE
- Timestamps
- Millisecond observation time; venue-specific source stamps
- Underlying price
- Nearest preceding Binance reference when available
- History
- 7/30/90-day windows + full archive (Desk)
- Delivery
- REST history + normalized live WebSocket frames
- Resolution
- Raw stored observations or ?interval= 1s–1d
What the MCP server gives your agent
Keyless to start, your key for everything
Add the endpoint with no key and your agent is on the demo tier — BTC Kalshi markets, top-of-book — so a one-click install works the moment it lands. Add a DepthFeed API key (Authorization: Bearer df_…) and the same session unlocks all 7 assets, Kalshi, Limitless, the full ladder, and deep history. Paid-only calls return a clear in-tool message, not an error, so the agent knows exactly what to upgrade.
Eleven read-only tools across three venues
Discovery and depth in one toolset: polymarket_search_markets / kalshi_search_markets / limitless_search_markets to find a market, then *_get_orderbook for the latest book and *_get_snapshots for history, plus screener for a cross-venue top-of-book view and whoami to read the session's plan. Every tool is read-only and exposes the same underlying DepthFeed records through tool-specific response envelopes.
Add it to Claude, Cursor, Codex, or ChatGPT
In Claude Code: `claude mcp add --transport http depthfeed https://api.depthfeed.com/mcp`. In Cursor, VS Code, Claude Desktop, ChatGPT (Developer Mode), and OpenAI Codex, add the same URL as a custom/remote MCP server. Because it is a standard remote MCP endpoint, any MCP-capable client works — copy-paste snippets and one-click buttons are in the docs.
Inspect the data before integrating
Start with a bounded sample, verify the fields and timestamps, then choose the API or research workflow that matches the job.
Questions, answered.
It's a remote Model Context Protocol endpoint (https://api.depthfeed.com/mcp) that lets AI agents — Claude, Cursor, Codex, ChatGPT — pull Kalshi order-book depth and price data as tools, without writing API code. The agent calls tools like kalshi_get_orderbook; DepthFeed returns the same JSON the REST API serves.