
Currents
Currents records every Playwright (and Cypress) test run in CI with traces, screenshots and history, detects flaky tests, orchestrates specs across machines, and gives Claude Code, Codex and Cursor the same record over an MCP server. From $49/month after a 14-day trial.
About Currents
What Currents is
Currents is a hosted dashboard and orchestration service for Playwright test suites that run in CI. Your tests keep running on your own CI machines, as they do today; a Currents reporter streams each result to the dashboard as it happens, and Currents orchestration tells each machine which spec file to run next. On top of that record it builds flaky-test detection, analytics, pull-request checks, Slack alerts and quarantine rules. Cypress suites are also supported side by side, which is how many of its customers migrated to Playwright one suite at a time.
The November 2026 launch, under the tagline "Currents for Claude & Codex to see all your tests", is about coding agents. A hosted MCP server and plugin listings let Claude Code, Codex, Cursor and other MCP clients read the same history a person sees in the dashboard: the error, the steps, the Playwright trace and the test's past results. The pitch is that an agent starts every session from zero, while Currents keeps months of runs, flakes and failures, so the agent can tell a flaky test from a real regression and fix the cause instead of the symptom.
Currents started as a side project and describes itself as a remote-first team; its site copyright runs from 2021. The founder and CEO is Andrew Goldis, and the company is backed by TinySeed, the accelerator for bootstrapped SaaS companies, and says it prioritises long-term sustainability over hyper-growth. Customers quoted on its site include Mezmo (LogDNA), Infraspeak, Housing Anywhere, Juvare, Crunchbase and Runa.
What it does
- Test evidence in one place. Every attempt keeps its Playwright trace, screenshots, video, visual diffs and console and network logs next to the error, so a failure opens in one click instead of digging through CI artifacts.
- Flaky test detection. Tests that pass on retry are flagged, and each test's flakiness rate and history is tracked across branches. AI classification groups errors by category, action and target across runs.
- Quarantine, skip and tag rules (Currents Actions). Rules quarantine, skip or tag tests without a code change; quarantined tests still run but cannot fail the build, and rules can expire.
- Smart orchestration. Instead of Playwright's static
--shardsplit, spec files are handed out from a live queue ordered by historical duration so all machines finish at about the same time. The company says this makes runs up to 50% faster, and that spot-instance support (a reclaimed machine's specs move to a healthy one) cuts CI cost by up to 70%. Fail-fast, risky-tests-first and rerun-only-failed modes reduce wasted CI time. - Agent access. A hosted MCP server (browser sign-in), a local npm MCP server, agent skills, and VS Code and Cursor extensions. Agents can fetch a failure's context, create share links that open without a Currents login, quarantine a test, file a Jira or Linear issue, or record before-and-after browser evidence of a fix and attach it to the pull request.
- Integrations. GitHub, GitLab and Bitbucket status checks and comments, Slack and Microsoft Teams alerts, Jira and Linear, a REST API and webhooks.
What we checked: the reporter CLI and the MCP server
On November 21, 2026 we installed the two packages from npm in a clean Node 22 sandbox, without a Currents account: @currents/[email protected] (MIT, published November 20, 2026) and @currents/[email protected] (Apache-2.0, published the same day; the source is on GitHub as currents-dev/currents-mcp).
$ npx pwc --version
2.12.0
$ npx pwc --help
Usage: pwc [options] [playwright arguments and flags]
--ci-build-id <id> the unique run identifier (env: CURRENTS_CI_BUILD_ID)
-k, --key <record-key> your secret Record Key (env: CURRENTS_RECORD_KEY)
-p, --project-id <project> (env: CURRENTS_PROJECT_ID)
--pwc-cancel-after-failures <number | false>
--pwc-console-output <preset> none | summary | compact | detailed
...
pwc --key <record-key> --project-id <id> --ci-build-id <build-id> -- --workers 2 --shard 1/2
pwc is a wrapper around playwright test: everything after -- goes to Playwright unchanged, and the record key, project ID and build ID can come from environment variables, which is how you keep the key out of CI logs.
Then we started the local MCP server over stdio and spoke raw JSON-RPC to it. With no key it refuses to start (CURRENTS_API_KEY is not set); with a placeholder key it answers initialize and tools/list without calling the API:
$ printf '%s\n' '{"jsonrpc":"2.0","id":1,"method":"initialize",...}' \
'{"jsonrpc":"2.0","id":2,"method":"tools/list"}' \
| CURRENTS_API_KEY=placeholder node node_modules/@currents/mcp/dist/index.mjs
serverInfo: currents 2.8.0, protocolVersion 2025-06-18
capabilities: tools, resources, prompts, extensions: io.modelcontextprotocol/skills
tools: 44 (26 marked readOnlyHint, 8 marked destructiveHint)
currents-get-context "Call this first to find out why tests failed in CI..."
currents-get-test-evidence screenshots, videos, traces, attachments
currents-create-action quarantine / skip / tag rules
currents-delete-run permanent deletion [destructive]
currents-create-share-link public link, no sign-in needed
...
What we took from it:
- It is a full API surface, not a read-only viewer. 18 of the 44 tools write: creating and deleting actions, cancelling and resetting runs, deleting runs, creating webhooks and public share links, and filing Jira issues. The server marks eight of them destructive, which MCP clients use to decide when to ask before calling.
- Scope the key. The documentation maps each tool to one permission (for example
results:read,actions:write,runs:write,shares:write). The hosted server lists only the tools a connection's permissions cover; the local server lists every tool and lets the API key refuse what it does not allow. An agent that only needs to debug failures needs read scopes alone. - The hosted endpoint requires auth. An unauthenticated POST to
https://mcp.currents.dev/mcpreturned HTTP 401 when we tried it. - Weight. The two packages pulled in 95 MB of
node_modulestogether; the MCP server alone has five direct dependencies (the official MCP SDK, commander, pino, pino-pretty and Zod 4) and needs Node 20 or newer.
Pricing (checked November 21, 2026)
| Plan | Price | Included | Notes |
|---|---|---|---|
| Scale | $49/month (10K test results); annual billing gives 2 months free | 10,000 test results a month, then $4.90 per 1,000 | 50 users, unlimited projects, AI fixes, orchestration, flaky detection, analytics, Actions, integrations, SOC 2 Type 2 |
| Business | $99/month (10K test results) | 10,000 test results a month, then $9.90 per 1,000 | Everything in Scale plus SAML SSO, SCIM provisioning, priority support |
| Enterprise | Custom | Any volume | Custom terms and retention, SLA, data redaction, self-hosting, bring-your-own S3-compatible storage, Slack support |
Larger monthly volumes (20K up to 500K results) are chosen on a slider; the pricing page's own example puts Scale at 30K results at $124 a month. There is a 14-day free trial but no free plan. Overage can be capped or switched off. Retention on Scale and Business: artifacts of passed tests 7 days, artifacts of failed, flaky or quarantined tests 21 days, test results 90 days, test details 30 days, reports and analytics 365 days.
Worked example: what a mid-sized suite costs
Currents counts a test result per test per run, and its estimator says to multiply suite size by CI runs. Take a 450-test Playwright suite:
450 tests x 30 CI runs/month = 13,500 results
Scale 10K: $49 + 3.5K x $4.90/1K = $49 + $17.15 = $66.15/month
Business 10K: $99 + 3.5K x $9.90/1K = $99 + $34.65 = $133.65/month
Same suite on every pull request: 450 x 200 runs = 90,000 results
Scale 10K + overage: $49 + 80K x $4.90/1K = $441/month (before picking a larger tier)
The second line is the scenario the agent pitch creates: every agent-written pull request is another CI run. At that volume, picking a bigger tier on the slider and using rerun-only-failed or fail-fast (both reduce results recorded) matters more than the base fee. Whether retries count as extra results is not stated on the pricing page; check before sizing a plan around a retry-heavy suite.
Currents vs the alternatives
| Currents | TestSprite | heal.dev | Appxiom | |
|---|---|---|---|---|
| Core job | Record, analyse and orchestrate the Playwright/Cypress tests you already have | AI agent that writes and runs E2E tests from plain language, PRDs or API specs | AI agent that explores the app in a sandbox and files bugs with a failing test | Production bug detection for mobile and web apps |
| Who writes the tests | You (or your coding agent) | TestSprite | heal.dev, as repro tests | No tests: SDK in the app |
| Runs where | Your CI machines | TestSprite's cloud | heal.dev's sandbox | Users' devices in production |
| Agent access | Hosted and local MCP server, 44 tools, skills, IDE extension | MCP server and CLI | Defects with tests | MCP server and agent setup skills |
| Pricing basis | Test results per month (from $49) | Credits (free plan, then paid tiers) | Paid | Per app (slabs) |
Microsoft's own Playwright Testing service and open-source reporters such as Allure and ReportPortal are the other usual comparisons; Currents publishes its own comparison pages against them, which are vendor-written.
Who it's for
Teams with a Playwright suite large enough that CI time, flaky tests and failure triage cost real hours; QA leads who want one history of every run across branches; teams moving from Cypress to Playwright; and teams whose pull requests are increasingly written by coding agents and who want those agents to read test results and fix failures with evidence rather than guess.
Limits and trade-offs
- No free tier. After the 14-day trial the cheapest plan is $49 a month; small projects can get by with Playwright's built-in HTML report and trace viewer.
- Volume-based billing. Cost scales with suite size times CI runs, so more frequent runs (including agent pull requests) raise the bill directly; set overage limits.
- Short artifact retention. Traces and videos of passing tests are kept 7 days and of failures 21 days on Scale and Business; longer history needs Enterprise.
- SSO costs double. SAML SSO and SCIM are on Business, at twice the Scale base and overage rates; self-hosting and bring-your-own storage are Enterprise only.
- Write access for agents. The MCP server can delete runs and webhooks and create public share links; give agents a key with read scopes unless you want them to change things.
- Playwright first. Cypress is supported, and other frameworks are listed in the docs, but the agent and AI features are built around Playwright traces.
Related on TechLogHub
- CI/CD and the GitHub Actions CI/CD pipeline tutorial.
- Playwright vs Cypress, the migration many Currents customers made.
- MCP, AI agent and observability.
- Webhooks and REST APIs, which Currents uses to export results.
- TestSprite, heal.dev and Appxiom.
- The Git cheat sheet and Docker cheat sheet for CI work.
Value & Audience
Value Proposition
One record of every CI test run, with evidence and history, that both people and coding agents can read to separate flaky tests from real failures, fix them and speed up CI.
Problem Solved
Playwright failures in CI are slow to triage from scattered artifacts, flaky tests erode trust in the suite, static sharding wastes CI time, and coding agents fixing tests have no history of past runs.
Target Audience
Tech Stack
Social & Links
Loading ratings...
Loading comments...
Gallery

Feedback & Roadmap
Currents FAQ
Common questions about Currents's features, pricing, and use cases
Is Currents free to use?
No. Currents is a paid product, starting at Scale - $49/month - 10K test results/month then $4.90 per 1K, 50 users, unlimited projects, AI fixes, orchestration, flaky detection, SOC 2 Type 2. Plans listed: Scale - $49/month - 10K test results/month then $4.90 per 1K, 50 users, unlimited projects, AI fixes, orchestration, flaky detection, SOC 2 Type 2; Business - $99/month - 10K test results/month then $9.90 per 1K, adds SAML SSO, SCIM, priority support; Enterprise - custom - any volume, custom retention, SLA, data redaction, self-hosting, bring-your-own storage.
What are some alternatives to Currents?
Popular Testing & QA alternatives to Currents include TheySaid, heal.dev and BotGauge.
Currents appears in
You May Also Like
TheySaid
TheySaid is a user-testing platform that runs usability tests with synthetic AI testers or real people and delivers findings and fix prompts inside tools like Claude Code, Cursor and Jira.
View Productheal.dev
heal.dev is an AI agent that runs an application in a production-like sandbox to find bugs, performance regressions and runtime security issues. Each defect it files comes with a failing test that reproduces it.
View Product
BotGauge
BotGauge is an AI agent reliability platform that red-teams agents with adaptive attacks, traces every run, turns failures into evaluations and policies, and monitors agents in production; the company also runs an AI-led managed QA service.
View ProductTestifly
Testifly is an autonomous AI QA service for web apps: you give it a URL, and it discovers your real user flows, turns them into end-to-end tests, runs them on every deploy and on a schedule, heals them when the UI changes, and explains failures with root-cause analysis and session video.
View ProductOne email a week
TechLogHub tracks 920 tools and 614 open-source projects. Get the week's new launches, release roundups and open-source picks — nothing else.
Get the TechLogHub digest
One email a week: new and trending developer tools, fresh comparisons, and what shipped. Unsubscribe in one click.