Crypto MCP server: 42 tools for Claude, Cursor, and any MCP client.

MCP (Model Context Protocol) is an open standard that allows AI assistants to connect to external data sources and tools. The Sharpe MCP server exposes 42 tools that give AI agents like Claude and Cursor direct access to real-time crypto market data. When you ask Claude a question like 'What are the highest funding rates right now?', the AI can call the Sharpe MCP server's get_funding_rates tool to fetch live data and answer with current information rather than relying on training data.

For most tools, no. Most MCP tools work without an API key by falling back to free public endpoints, so you can add npx -y @sharpe-terminal/mcp-server or the hosted endpoint to your AI tool's configuration and start querying with no signup. Ten tools need a SHARPE_API_KEY (the free tier works): get_market_overview, market_briefing, analyze_coin, find_opportunities, get_exchange_listings, get_listing_events, get_listing_exchanges, get_recent_listings, get_stablecoins and get_derivatives_overview. A key also gives higher rate limits.

The Sharpe MCP server is compatible with Claude Desktop, Claude Code, Cursor, and any MCP-compatible client. Configuration is one entry in the tool's MCP settings: the hosted endpoint's URL, or the npx -y @sharpe-terminal/mcp-server command. The npm package speaks stdio and the hosted endpoint speaks Streamable HTTP, the two standard MCP transports, so as the MCP ecosystem grows, any new MCP-compatible AI tool will work with the Sharpe server without modification.

The three composite tools are analyze_coin (deep dive on a single coin combining funding, futures, and market data), market_briefing (market overview with top narratives, funding highlights, and arbitrage opportunities), and find_opportunities (scans all data products for high-conviction setups). Use a composite tool for a broad question. Use individual tools like get_funding_rates or get_futures_data when you need specific data points.

Run npx -y @sharpe-terminal/mcp-server: npx downloads the published npm package and runs it over stdio without a global install. For Claude Desktop, add a server to claude_desktop_config.json whose command is npx with the arguments -y @sharpe-terminal/mcp-server. For Claude Code, run claude mcp add sharpe -- npx -y @sharpe-terminal/mcp-server. For Cursor, add the same command to .cursor/mcp.json. Clients that take a URL, including Claude.ai and ChatGPT custom connectors, can use the hosted endpoint https://www.sharpe.ai/api/mcp instead. The server starts when the AI tool launches, so no separate server process or port configuration is needed.

The MCP server wraps the same REST API but is optimized for AI consumption. Differences include: smart defaults so AI agents get useful results without specifying every parameter, composite tools that synthesize multiple API endpoints into single responses, response formatting optimized for LLM context windows, and automatic truncation of large datasets. The REST API is better for programmatic integration into trading systems, while the MCP server is designed for conversational AI workflows.

Data freshness through the MCP server matches the underlying Sharpe data products: current funding rates refresh every 5 minutes, futures datasets every 10 minutes to 6 hours depending on the metric, narratives and ecosystems every 4 hours, and news every 15 minutes. The MCP server fetches live data on each tool call, with no extra caching layer in between.

Yes. The MCP server is free to install and many tools work without an API key by falling back to free public endpoints. Authenticated-only `/api/v1` tools require `SHARPE_API_KEY`; add one for complete coverage, higher rate limits, and automated workflows.

MCP is an open protocol specification introduced by Anthropic in late 2024 that standardizes how AI assistants connect to external data sources, tools, and services. An MCP server exposes tools (functions the AI can call), resources (data the AI can read), and prompts (templates the AI can use). MCP uses JSON-RPC over stdio or HTTP transport. The protocol is tool-agnostic: any MCP-compatible client (Claude Desktop, Claude Code, Cursor, Zed) can connect to any MCP server. That decouples AI assistants from specific integrations, the way USB decoupled peripherals from operating systems.

The MCP server is built for conversational AI work such as interactive research, market briefings, and agent-driven analysis, not for high-frequency automated trading. For programmatic trading systems, Sharpe's REST API is the better choice: lower latency, deterministic schemas, and direct HTTP access without the MCP transport overhead. MCP fits semi-automated workflows where an agent periodically checks market conditions, or Claude Code users querying live market data during development.