agnostic-ai v0.71.0: user MCP servers, default models, and import --global

Agnostic-ai released v0.71.0, adding user-level MCP server installation for Claude Code, Codex, Cursor, Copilot, Gemini, Qoder, OpenHands, and Augment, plus a settings spec that sets default model and effort per tool. The release also adds `agnostic-ai import --global`, which converts existing model, effort, and MCP server settings into home specs so the next sync makes no changes. The project warns that current OpenHands releases ignore MCP servers in a project `config.toml` [mcp] section as legacy V0, so users should copy those specs into `~/.agnostic-ai/mcps/` and run `agnostic-ai sync --global`, which now writes `~/.openhands/mcp.json`.

What to do 1. If a project syncs MCP servers to OpenHands, copy those MCP specs into ~/.agnostic-ai/mcps/ and run agnostic-ai sync --global : current OpenHands releases ignore the project config.toml one copied file per server . Nothing else needs action. What changed - MCP specs in ~/.agnostic-ai/mcps/ install into the user MCP files of Claude Code, Codex, Cursor, Copilot, Gemini, Qoder, OpenHands, and Augment 1242, 1246, 1252, 1253 . - A settings spec in the home sets the default model and effort for Claude Code, Codex, Copilot, Qoder, and Gemini, and model now takes a per-tool map 1240 . - agnostic-ai import --global turns the model, effort, and MCP servers your tools already hold into home specs, so the next sync changes nothing 1243 . Full details are in the GitHub release notes https://github.com/Chemaclass/agnostic-ai/releases/tag/v0.71.0 . Upstream CLI and model news - OpenHands, current releases docs checked 2026-09-28 : MCP servers in a config.toml mcp section are legacy V0 and ignored, so servers synced there never connected. sync --global now writes ~/.openhands/mcp.json , and the project output is deprecated. State: supported. MCP settings https://docs.openhands.dev/openhands/usage/settings/mcp-settings