The tools cover the lookups a Web3 conversation actually needs — price, TVL, yields — with nothing to configure, because there is no wallet and no write path anywhere in the surface. The 42 agent definitions sitting next to them are plain JSON in a standard format, so they can be pointed at any platform that reads an agent index rather than being tied to this one.
Two things in one repository. The MCP surface is a hosted HTTP endpoint returning live market and DeFi data: coin prices, market overview, trending, protocol and chain TVL, and yield opportunities. Alongside it, the project publishes 42 AI agent definitions for DeFi, portfolio management, trading and Web3 education as a machine-readable JSON index in 18 languages, usable by any platform that reads agent indexes.
- Prices for named coins against chosen currencies, and a market overview with a result limit — `get_price`, `get_market_overview`
- What is moving and what matches a search, plus per-coin detail and global market statistics — `get_trending`, `search_coins`, `get_coin_detail`, `get_global_stats`
- DeFi protocols ranked by TVL and the detail behind one of them — `get_defi_protocols`, `get_protocol_detail`
- Total value locked by chain, and current yield opportunities — `get_chain_tvl`, `get_yield_opportunities`
- 42 agent definitions as readable JSON, retrievable per locale from the project's index, with a manifest for indexing — no hidden behaviour, and no platform lock-in
Nothing to authorise for the hosted path: point the client at `https://modelcontextprotocol.name/mcp/defi-agents` with type `http`. There is also a published npm package, `@nirholas/ai-agents-library`, registered as `io.github.nirholas/defi-agents` with a stdio transport and no required environment variables. The agent index itself is fetched over plain HTTPS as JSON, per locale, and the project is served from a CDN.
One command — npx -y mcp-remote https://modelcontextprotocol.name/mcp/defi-agents