Agents
MCP compliance tracker scores five coding agents at 56% and finds the Tasks spec implemented by none
ismcpgoodyet.com tracks Codex, Cursor, Claude Code, Grok and OpenCode against the MCP specification, currently scoring the field at 56% overall. Tasks, the spec feature for long-running work, is absent across all five; the stateless spec revision dated 2026-07-28 is only partially supported by Codex and Claude Code; Cursor has no tools/list pagination at all; and skill:// resources over MCP are shipping nowhere, with only Codex partially importing snapshots. In-client OAuth, per-tool filtering and approval policies are universal, so the gaps sit squarely in the newer async and skill-distribution parts of the protocol builders are being told to design against.
↳ Follow the thread