Bitcoin DeFi has no benchmark lending rate. SBOR starts with Stacks, where the market is: one borrow and one supply rate per currency, read from lending contract state and published daily. This server exposes it as tools for Claude, Cowork, Cursor or any MCP client.

Quick check

npx -y sbor-mcp

It starts and waits silently. That is success. Ctrl+C to exit.

Install

{
  "mcpServers": {
    "sbor": {
      "command": "npx",
      "args": ["-y", "sbor-mcp"]
    }
  }
}

Claude Desktop: add that to claude_desktop_config.json and restart.

Or connect by address, with nothing to install

https://mcp.sbor.xyz/mcp

Streamable HTTP, no key, no sign-in. Works as a custom connector in Claude on the web and in any MCP client that accepts a URL. The remote server is generated from this one, so both give the same answers.

Tools

Tool	What it answers
`get_rate`	What does capital cost on Stacks right now, and what does borrowing USDC against bitcoin cost
`compare_rate`	Is this offer above or below the market, and by how much
`list_markets`	Which venues make up the rate, with utilization and depth
`get_history`	How has the rate moved
`compare_chains`	How Stacks compares with Ethereum, Base, Hyperliquid, Solana and SOFR
`get_methodology`	How the number is built, and what it excludes

All six tools are read-only.

Benchmarks

SBOR-USD, SBOR-BTC and SBOR-STX are the Stacks indices. BTC-COLLATERAL-USDC is a reference, not an SBOR index: what it costs to borrow USDC against bitcoin wrapped by a custodian (cbBTC, WBTC), from the Morpho markets on Base and Ethereum whose only collateral is that bitcoin. Use it to check a USDC loan against bitcoin on those chains.

Recommended use

Use SBOR to stop, never to start. Before borrowing, call compare_rate. If the offer is more than 50 basis points above the benchmark, stop and ask a human. A wrong reading under that rule costs a pause, not a trade.

Behavior worth knowing

It refuses rather than guesses. compare_rate returns an error, not a verdict, when the data is more than 48 hours old, when the fixing timestamp cannot be read, when the index is not published, or when a rate is missing. An error means there is no trustworthy answer, not that the rate is bad. Fall back to your own logic.

It checks units. Rates are percentages: 4.2 means 4.2%. Anything outside 0 to 100 is rejected. Below 0.5 is answered, because rates that low genuinely occur, but with a warning that you may have passed a fraction.

It never averages across a methodology change. get_history gives a mean per methodology version when a window spans more than one.

It never guesses when SBOR is down. If SBOR is unreachable or slower than 10 seconds, the tool says so rather than substituting an estimate.

It reports omissions. When a market cannot be read, SBOR omits the index rather than publishing a figure that is not real. The tool explains that instead of returning nothing.

It surfaces concentration. An index covering one venue is a reading of that venue, not a market average, and get_rate says so.

It reads the same public endpoints as everyone else. No key, no state, no writes, no telemetry. Set SBOR_BASE to point at a different host.