CircleCI
Continuous integration and delivery platform for building, testing and deploying software
CircleCI ships an MCP server and auto-rollback deployments, moving well past pure CI.
◆Recent moves
- 1d ago
Machine Runner Orchestrator Release 1.0.0
⚡ SPARKMachine Runner Orchestrator reaching GA closes the self-hosted scaling story CircleCI has been building: one deployment now handles CPU, GPU, and ARM workloads, ships Prometheus metrics, and runs on GKE, AKS, EKS, or any KubeVirt cluster. The breaking rename from runner-provisioner requires a full uninstall/reinstall for existing deployments.
View source ↗ - 1d ago
Xcode 27.0 Available
Standard Xcode 27.0 image availability — expected maintenance cadence with no user-visible capability changes beyond the updated toolchain.
View source ↗ - 2d ago
CircleCI is now available on Cursor Origin, in beta
CircleCI integrates with Cursor Origin — Cursor's AI-native VCS — in beta, letting teams run existing .circleci/config.yml workflows against Origin repositories with no config changes. It's the first CI integration for Cursor's emerging VCS and signals CircleCI's intent to be present wherever AI-native development workflows land.
View source ↗ - 2d ago
Hosted MCP server now available
⚡ SPARKThe hosted MCP server is the most direct expression of CircleCI's agentic trajectory: AI assistants now have 24 tools to query, diagnose, and act on CI pipelines without leaving the chat. The surface covers the full failure-diagnosis loop — from listing runs to reading logs to rerunnning workflows — with no installation required.
View source ↗ - 2d ago
Xcode 27.0 RC Available
Xcode 27.0 RC image available ahead of GA — preparatory release for teams tracking Apple developer betas, no capability changes.
View source ↗ - 5d ago
Build tool cache beta: up to 90% faster clean builds for Gradle, Bazel, and Turborepo
The `with_tool_cache` step removes manual save/restore cache pairs for Gradle, Bazel, Turborepo, and Xcode workloads, delivering up to 90% faster clean builds without additional cost. This fills a gap where teams were hand-rolling cache configurations with inconsistent results across tool versions.
View source ↗