If youâve been using LLMs to write or edit your Bitrise configuration, youâve probably hit the same wall: the AI confidently generates a bitrise.yml that references steps that donât exist, picks a stack with the wrong Xcode version, or produces a config that looks valid but breaks on the first run. Weâve fixed that.
Whatâs new
Step search
The Bitrise MCP server can now search the live step library directly. When you ask your LLM to add a deployment step or set up code signing, it queries real step IDs, versions, and inputs â rather than hallucinating them. No more spending 20 minutes debugging a config because the AI invented a step name.
Stack search
Your AI assistant can now look up available build stacks and machine types before writing your config. Ask it âwhat stacks support Xcode 16?â and it will tell you â accurately. For teams on dedicated infrastructure or AWS, passing an app_slug unlocks validation against your specific machine types and license pools.
YAML validation
Before any changes are applied, the MCP server can now validate your bitrise.yml for both syntax and semantic correctness. This means your AI workflow catches errors at generation time, not at runtime. Think of it as a CI check for your CI config.
Tool annotations
Weâve added structured annotations to all 60+ MCP tools, classifying each one as readonly, destructive, idempotent, or open-world. In practice this means your AI agent now knows to freely call read operations, but pause and ask for confirmation before overwriting your pipeline config. Fewer surprises, more control.
Getting started
If you havenât set up the Bitrise MCP yet, head to the Bitrise MCP repository to get started.
Weâd love to hear how youâre using AI with your Bitrise setup â drop us a note or open an issue on GitHub.