April 2026
MCP-first pivot — runbooks for agentic coding
Repositioned from a single-IDE plugin to an MCP server that plugs into the team's AI host — Claude Code and Cursor. The admin portal, templates, checkpoints, and audit trail become the team layer above whatever agent each engineer uses.
new
mcp server
positioning
/mcp/* server shipped — JSON-RPC tools (list_runbooks, start_run, request_checkpoint, create_pr, …) over HTTP + SSE
- Plugin API keys:
rail_pk_* format, sha256-hashed, revocable via the new admin plugin-keys page
- CheckpointRequest model + admin approval UI so MCP-driven flows pause for human review at phase boundaries
- Pricing simplified: free forever up to 3 engineers, $9/seat from seat 4 (all seats billed); single interactive calculator on the pricing page
- Mobile scope expanded beyond Android + KMP to include iOS, Flutter, React Native as first-class targets (host agents run native builds locally)
- PluginTelemetry pipeline lands every agent session in Postgres — remote diagnosis without user-submitted logs
- New
/vs.html positioning page: Linear vs Factory.ai vs RailRun, gap matrix, MCP constellation