CircleCI MCP

The CircleCI MCP server gives an AI client direct access to build data so it can explain a red pipeline instead of speculating. get_build_failure_logs returns failure logs with step by step execution detail, get_latest_pipeline_status reports workflow status, duration and timestamps, and get_job_test_results returns test metadata, results and timings. find_flaky_tests analyzes execution history to identify tests that pass and fail without code changes. config_helper validates a .circleci/config.yml for syntax and semantic errors before it is pushed. list_followed_projects, list_artifacts and list_component_versions cover project discovery, build artifacts with download URLs, and deployed component versions with commit information. Write operations are limited and explicit: run_pipeline triggers a pipeline, rerun_workflow reruns a workflow from the start or from the failed job, and run_rollback_pipeline executes a rollback across environments. Cost and capacity work is covered by download_usage_api_data, which exports usage metrics as CSV, and find_underused_resource_classes, which reads that export to flag jobs whose average or peak CPU and memory sit below a threshold and could run on a smaller resource class. Two access paths exist side by side. A hosted server accepts OAuth or a personal API token with no local install, and the CircleCI CLI ships an MCP mode that reuses existing CLI authentication and exposes the wider CLI surface, including config authoring and org management. It is aimed at developers and platform engineers debugging CI from a coding assistant.