Summary

The test data an agent needs to fix a failure, next to the controls that stop a flaky test blocking everyone while it works.

Most CI integrations stop at "here is the run". This one goes as deep as a fix actually needs: failure context formatted for a model, a stable test signature so history survives a rename, and flakiness ranked across the whole suite. The other half is action — quarantine rules, run resets and cancels, and a Jira issue created from the failing test — so an agent can change what CI does rather than only read what it did.

What it is

A server over the Currents dashboard covering CI test data end to end: projects and runs, individual test results and their history, spec-file and test performance, aggregated error metrics, the rules that skip or quarantine tests, Jira issue creation, and webhooks.

What you get

Runs found and read: currents-get-runs filters by branch, tags with AND/OR, status, author and date range; currents-find-run returns the most recent completed run for a ciBuildId or a branch-and-tag combination; currents-get-run-details opens one
currents-get-context returns test-failure context shaped for debugging at run, instance or test level, as JSON or Markdown, with a detail level and pagination
currents-get-tests-signatures builds a test's stable signature from project, spec path and title, and currents-get-test-results returns that signature's history filtered by branch, tags, author, status and flakiness
Trends across the suite: currents-get-tests-performance and currents-get-spec-files-performance rank by flakiness, failure rate, duration or executions, and currents-get-errors-explorer aggregates errors by message, category and target
Actions — the rules that automatically skip, quarantine or tag tests — created, updated, enabled, disabled and archived, with currents-list-affected-tests and currents-get-affected-executions showing what each rule has actually caught
Run control: cancel a run in progress, cancel one by GitHub Actions workflow run ID and attempt number, reset failed spec files for re-execution, or delete a run and its data outright
Jira straight from a failing test: list the projects and issue types the organisation integration offers, then create an issue from a run test or link an existing one
Webhooks per project listed, created, updated and deleted, with the run events that fire them
currents-get-project-insights returns run and test metrics over a date range at a chosen resolution, and currents-list-pull-requests groups runs into pull-request cards

Requirements

A Currents API key in CURRENTS_API_KEY, launched over stdio with npx as @currents/mcp. The same server also runs as a hosted Streamable HTTP endpoint at https://mcp.currents.dev/mcp for remote connectors; that endpoint performs no authentication of its own, so each request must carry your Currents API key as a Bearer token, which it passes through to the Currents API. The server publishes its agent skills as MCP resources under skill://currents/, so a connected agent can read them without a local copy.

Setup effort

One command plus a key — npx -y @currents/mcp, then supply credentials