v0.4.0 - The MCP-recorder-first rewrite

AutomatiQ is now an MCP server that records your browser sessions and compiles them into agent-ready artifacts. Browse a site normally while AutomatiQ captures every request, WebSocket frame, and user action - then your agent reads the resulting workspace and writes the HTTP-replay script. No browser at runtime.

Install: pip install automatiq - or paste the Quickstart block from the README into your agent and let it do everything.

Highlights

• Five-tool MCP surface over stdio: start_recording, stop_recording, wait_for_completion, get_status, annotate_user_interactions
• Deterministic per-session folders under automatiq_sessions/, each shipping a generated README that documents every artifact
• Vision analysis on by default (with a key): per-action clips annotated with intent and outcome, re-runnable later via annotate_user_interactions
• Config-only keys: recorder_api_key in ~/.automatiq/config.toml - provider env vars are never consulted
• Telemetry v2: server_started, tool_called (every tool invocation with duration and errors), annotate_run
• Python 3.11+; Brave-only recorder with managed download

Breaking

The CLI product (record/run/agent/resume/feedback), the LLM investigator agent, and the IPython sandbox are gone - archived on the legacy/v0.3.x branch.

See CHANGELOG.md for the full entry.