blackswan

BlackSwan MCP Server provides two tools via the MCP interface: Flare and Core. Flare offers real-time precursor detection over a 15-minute signal window to answer: Is something happening right now? Core synthesizes the overall risk environment using a 60-minute signal window to answer: What's the overall risk environment? You can query the MCP endpoint at the root / mcp path, or use the REST API endpoints /api/flare and /api/core to fetch structured results. For Claude or other clients, you can add the server to your configuration as shown in the README example, and then ask natural-language questions like: What does Flare say about current crypto risk? The responses include agent identifiers (flare or core), timestamps, data ages, assessments, and structured fields such as signals for Flare or key factors and environment for Core. Practically, you’ll leverage these tools to monitor current risk conditions and understand drivers behind risk shifts, enabling proactive decision-making. To consume programmatically, call the /api/flare endpoint for current precursors or the /api/core endpoint for the environment assessment, parse the agent, data_age, timestamp, environment (for Core), and the accompanying details.

Prerequisites:

• A hosting environment with outbound Internet access
• TLS certificate and domain configured for the MCP endpoint at mcp.blackswan.wtf (as used in examples)
• Optional: a client (CLI or code) capable of issuing HTTP requests to the MCP endpoints

Step-by-step installation (illustrative, adapt to your environment):

1. Decide your deployment method (HTTP-based MCP server exposed at /mcp). Ensure you have a reverse proxy/TLS in place to serve /mcp over HTTPS.
2. Deploy the server that implements the MCP endpoints (mcp at /mcp). If you are reusing the upstream BlackSwan MCP server as shown, you only need network access to https://mcp.blackswan.wtf/mcp and the REST endpoints. If you are hosting yourself, implement an HTTP-based MCP interface that proxies requests to /api/flare and /api/core as described in the README.
3. Configure Claude or any MCP client to point to the server. Example Claude config:

{"mcpServers": {"blackswan": {"type": "http", "url": "https://mcp.blackswan.wtf/mcp"}}}

4. Validate endpoints:

• curl -s https://mcp.blackswan.wtf/api/flare
• curl -s https://mcp.blackswan.wtf/api/core

5. Monitor logs and ensure the server produces timely agent outputs (flare/core) with the required fields as specified in the README. If using your own deployment, implement error handling for HTTP status codes 200, 502, 503, 500 as documented, and return a JSON payload with an error message on non-200 responses.

Tips and considerations:

• The Blaze/Flare and Core outputs include data_age and timestamps; ensure your system clock is synchronized (NTP) to avoid timestamp drift.
• If you encounter 502 or 503 errors, it typically indicates upstream agent output issues or startup delay; retry after a short interval.
• When configuring clients, use the exact REST endpoints: /api/flare for Flare results and /api/core for Core results. Structure of the responses will follow the field definitions in the README (e.g., agent, data_age, timestamp, environment, assessment, key_factors, sources_used, data_freshness).
• No API key is required for the public MCP endpoint, but consider adding access controls if hosting privately.
• If you publish a Claude config or a client config, keep the URL in TLS-enabled form and validate certificates to prevent man-in-the-middle issues.