Steyble

Steyble DeFi infrastructure for AI agents, exposed over the Model Context Protocol. Three servers, one shared security layer: every tool is classified, permission-checked, rate-limited, output-sanitized (secret scan), and audited (hashed args, never raw).

Package	Server	Tools	Needs API key
@steyble/mcp-server	Platform API (swap, bridge, stake, predict, perps, trading, yield)	21	yes (b2b_live_...)
@steyble/mcp-developer	Developer Portal (examples, validation, SDK snippets)	7	no
@steyble/mcp-public	Public / Education (Web3 + MPC concepts, public API facts)	9	no

Quick install (Cursor)

One-click installs (or see each server's README for manual mcp.json config):

• Platform API: Add to Cursor
• Developer Portal: Add to Cursor
• Public / Education: Add to Cursor

Security model

• Read-only by design — no tool signs, broadcasts, or executes transactions. Yield enter/exit tools return unsigned payloads for you to review and sign.
• Defense in depth — every tool runs through a shared defineTool pipeline: classification → permission policy → rate limit → input validation (zod) → output sanitization (secret patterns scrubbed) → audit (sha-256 arg hash only).
• Secrets never leave your machine — API keys go only to the Steyble API; CI runs a secret scanner over all built artifacts.

Develop

npm install
npm run verify          # build + secret-scan
npm run start:platform  # STEYBLE_API_KEY=... required

License

MIT — see LICENSE.