pi-mcp-adapter uses Jev for semantic tool search to cut tokens
pi-mcp-adapter adds async MCP Tasks and semantic tool search via Jev, sending full schemas only for confident matches to use up to 5× fewer tokens.
Just shipped pi-mcp-adapter updates for making MCP work smoother: On servers that support MCP Tasks, long-running tools now run async: the agent keeps working, and the result arrives when the job finishes. Semantic tool search is sharper: mcp({ search, searchMode: "semantic" }) with Jev or Clef sends full schemas only for confident matches, so a search costs up to 5× fewer tokens. Also: mcp({ tool, args, extract }) uses Cloudflare's Clairic to return just the values the agent asked for, not the whole result.