Why we put Currents inside Claude and Codex
Currents started with one engineer and one test problem.
In 2019, Andrew was the director of engineering at a startup. His team had trouble scaling their browser tests. He built an open-source tool to fix it and called it sorry-cypress. Soon it helped many companies scale their Cypress tests to millions of test records. In 2021, Andrew turned it into a hosted product: Currents.
The idea was simple. Every test run in CI leaves evidence: errors, steps, traces, screenshots, videos and logs. Most of that evidence disappears when the CI machine shuts down. Currents keeps it. It also joins each result to the history of that test across runs and branches. With that record, you can answer the questions that matter when a test fails:
Did this test fail before this change?
Does it fail only in CI?
Did other tests fail with the same error?
Market shifted and we put our focus on Playwright. After a year of work with the Playwright community, our Playwright reporter reached 1.0 in April 2024. Today, hundreds of teams use Currents, from startups to Fortune 500 companies, and they run millions of tests on it.
Then agents changed where the work is.
Agents write a lot of code, and they write it fast. But every change still needs proof that it works. Someone has to run the tests, read the results, fix what broke and run them again. Writing code stopped being the slow part. Verification became the slow part.
We saw our customers hit this. An agent opens a PR, and CI goes red. The agent can't see CI, so it says "done" and moves on. A person opens the Currents dashboard, finds the failure and copies the error into the chat. Then the agent tries again. The record already had everything the agent needed. But the agent had no way to read it.
So in 2026, we built the agent side of Currents, one piece at a time:
July: an IDE extension for VS Code and Cursor, with a "Fix with agent" action.
August: "Fix with AI" opens a failure in Claude and Claude Code. Slack alerts got the same button.
September: shareable AI context. Add .md to a failure link, and an agent gets the full context as Markdown. Later that month came a hosted MCP server at mcp.currents.dev, with OAuth sign-in.
October: a Trace API, so an agent reads a Playwright trace in small pieces. In our example, that took 4 calls and about 1,600 tokens, against about 20,000 for every file in the trace. We also added a "Fix with AI" section to the GitHub PR comment.
Each piece helped, but one problem stayed. To connect an agent, you had to find our MCP docs, copy a server URL, create an API key and edit a config file. That is a lot of work before you see any value.
An MCP server only helps if people can find it and connect it from the tools where they already work.
So we put Currents where they already look: the Claude directory and the ChatGPT/Codex plugin directory. You find Currents, click connect and sign in. Your agent acts as you, with your Currents role. Nobody on your team gets data access until they sign in themselves.
That is why we're doing this launch.
On the surface, it is a listing and a sign-in button. Under it are seven years of test records and one simple idea: a failed test is only useful if someone reads it. Now that someone can be your agent.
People still make the calls that need judgment. Is it a bug? Should the agent fix it? Should it merge?
Currents gives the agent the evidence. You keep the decisions, and you stop copying stack traces between windows.


p/runevr
p/producthunt
p/google-cloud-platform
p/operum
p/murror
p/tabstack
p/dino-kids
p/any-video-to-text
p/seldon
p/rithub
p/jot-2-0
p/okapi-2
p/gatrends
p/appxiom
p/flamigo
p/dots-7
p/kodlo
p/replay-11
Replies
Be the first to reply
Have a question or a thought to share? Add a comment above to start the conversation.