v0.2.1 — listed in the official MCP Registry

Highlights

Baton is now in the official MCP Registry as io.github.ribato22/baton — discoverable by MCP clients and the directories that index it (mcp.so, PulseMCP, Glama).

Install it clone-free from any MCP client:

uvx --from "baton-orchestrator[mcp]" baton-mcp

• Adds a validated server.json + a GitHub Actions workflow that publishes it to the registry via OIDC on each release (no stored token), gated on the PyPI ownership marker.

pip install baton-orchestrator        # CLI + library (imports as `baton`)

See CHANGELOG.md.

v0.2.0 — installable MCP server + quality-first routing

Highlights

• Installable MCP server — run it clone-free from any MCP client:

  uvx --from "baton-orchestrator[mcp]" baton-mcp

  (or pip install "baton-orchestrator[mcp]" then baton-mcp). A starting-point server.json is included for MCP registries.
• Quality-first routing (new default) — each sub-task goes to the strongest model capable of it. --prefer cash_protect_quota restores subscription-quota right-sizing.
• Python 3.11+ support, Node-24 pinned Actions, secret scanning, and a Mermaid architecture diagram.

pip install baton-orchestrator        # CLI + library (imports as `baton`)
baton "your goal"

See CHANGELOG.md for the full 0.2.0 notes.

v0.1.0 — first public release

First public release of baton-orchestrator (imports as baton).

pip install baton-orchestrator

A from-scratch, cross-provider multi-model AI orchestration engine: a supervisor plans a task DAG, routes each sub-task to the cheapest capable model, runs it one-shot or in an agentic tool loop, then synthesizes a final answer — plus a CLI, a live Web UI, and an MCP server for IDE agents.

See CHANGELOG.md for the full 0.1.0 notes.