Suitest is an open-source, MCP-native QA platform. Generate, run, and publish end-to-end tests with video evidence. Self-hosted.
One command boots the whole platform on your laptop — dashboard, SQLite, and your IDE agent wired in. Scale up to Docker or Helm when the team does.
npx @suiflex/suitest onboardFull platform on your laptop: web dashboard + SQLite + IDE MCP wiring. Connect your workspace model in Settings before running tests. Node 18+ and uv.
npx @suiflex/suitest-mcp initDetects your IDE and framework, writes suitest.config.json, and registers the MCP server. Restart the IDE and prompt the agent. Node 18+, Python 3.11+.
git clone https://github.com/suiflex/suitest && cd suitest
cp .env.example .env
make docker-up
open http://localhost:3000Full platform: postgres, redis, minio, api, web, runner. App images are prebuilt on ghcr.io — make docker-up pulls them, no local build.
helm install suitest infra/helm/suitest \
-f infra/helm/suitest/values.yamlKubernetes install with an air-gapped configuration for cluster-hosted models.
git clone https://github.com/suiflex/suitest && cd suitest
make devDev setup with pnpm workspaces and uv.
One conversation in your IDE becomes a full test lifecycle.
Tell your IDE agent "test this app with Suitest". The MCP server analyzes your app or crawls a URL.
The run engine drives a real browser through every generated case. Deterministic, repeatable, CI-friendly.
Cases, runs, video recordings, and traces land in your dashboard. Failures come with evidence attached.
Evidence is recorded for humans. get_failure_context serializes it for the model that is going to repair the code.
## checkout-flow: FAIL at step 4/7
**Step 4**: click "#submit-btn"
**Error**: TimeoutError: locator not visible after 30s
**DOM at failure** (excerpt around selector):
<form id="checkout">
<button id="submit-button" disabled>Pay now</button>
</form>
**Console** (errors and warnings only):
[error] POST /api/cart 500 (Internal Server Error)
**Network** (non-2xx only):
POST /api/cart returned 500
**Evidence**: video, screenshot, traceSample payload. Two bugs in one read: the id issubmit-button, not submit-btn, and the cart API is returning 500.
run_tests comes back 9/10. Most tools stop here and hand you a red X.
get_failure_context returns a budgeted markdown bundle: the failed step, a DOM excerpt around the selector, console errors, and non-2xx requests. Under 8 KB, built for a context window.
Wrong selector and a 500 from the cart API, both visible in one payload. The agent edits the app, not just the test.
Deterministic replay confirms the fix. The run, video, and history publish to the dashboard.
A PRD-generated suite executing against the bundled demo app: API and browser steps and screenshots included. Connect your workspace model before replaying it.
make demo, then openlocalhost:3000.Features
Claude Code, Cursor, and Codex talk to Suitest through one MCP server. Analyze, generate, run, and publish without leaving the editor.
{
"mcpServers": {
"suitest": {
"command": "npx",
"args": ["-y", "@suiflex/suitest-mcp"]
}
}
}Test a URL with no repo access, generate from your source, or run the pytest, Vitest, or Jest suite you already have. One TCM, and every case says which one it was.
Each case ships with a video recording and traces. A failing test is a bug report you can watch.
Runs use the validated workspace model and remain fully replayable. Same input, same result, every time.
Cases, runs, steps, and history live in a dashboard your whole team can read, not in one developer's terminal.
Everything the agent needs to take a test suite from first crawl to published report, exposed as plain MCP tools.
Analyze a repo, generate cases, run them, read failures, publish.
Test any web app from a URL and credentials. No repo access needed.
Run the pytest, Vitest, or Jest suite already in the repo and publish it as cases, runs, and coverage.
Arguments, return envelopes, and examples for every tool live in theMCP tool reference.
Laptop, server, cluster, or nothing but a CI runner. Same engine, same config file.
A solo developer, one laptop
The full platform in one command: web dashboard, API on SQLite, run supervisor, and your IDE's MCP config. No Docker; connect a workspace model before running tests.
npx @suiflex/suitest onboardA team on its own server
The whole platform: Postgres, Redis, MinIO, API, web dashboard, and runner. Prebuilt images are pulled from ghcr.io. Your data never leaves your infrastructure.
git clone https://github.com/suiflex/suitest && cd suitest
cp .env.example .env && make docker-upPlatform teams, air-gapped environments
Helm chart for production, including cluster-hosted models for generation and diagnosis with zero egress.
helm install suitest infra/helm/suitest \
-f infra/helm/suitest/values.yamlProjects that just want a merge gate
The GitHub Action runs the lifecycle inside the workflow and reports to the PR. No Suitest server required at all.
- uses: suiflex/suitest/action@main
with:
config: suitest.config.jsonProvider location does not change the product. Validate the workspace model once, then Suitest uses it for MCP execution and AI workflows.
Sign in, organize workspaces, and manage test cases before connecting a model.
Save any supported hosted or self-hosted provider in Settings, then validate it once.
Run MCP tools, execute suites, generate tests, and diagnose failures with the same feature set.
One composite action runs the suite, upserts a single PR comment with per-failure excerpts, and gates the merge with its exit code.
name: e2e
on: pull_request
jobs:
suitest:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- run: npm ci
- run: npm run start & npx wait-on http://localhost:3000
- uses: suiflex/suitest/action@main
with:
config: suitest.config.json0 all passed1 test failure, PR check goes red2 infra error, no false verdictSetup, exit codes, and local preview with --dry-run in theCI guide.
Teams usually stitch together a case manager, a runner, and an AI testing service. Suitest is the union of those roles, and you host it.
| Capability | Suitest | TestSprite | Playwright MCP | TestRail |
|---|---|---|---|---|
| Open source, self-hosted | Yes | No | Yes | No |
| Manual TCM before connecting an LLM | Yes | No | Yes | Yes |
| Test management UI (cases, suites, history) | Yes | PartialPartial | No | Yes |
| Deterministic, replayable runs | Yes | PartialPartial | Yes | No |
| Agent generates and runs tests from the IDE | Yes | Yes | PartialPartial | No |
| Agent-readable failure context for auto-fix | Yes | No | No | No |
| Video and trace evidence per run | Yes | Yes | PartialPartial | No |
Capability summary based on public documentation, July 2026. Details and caveats in the FAQ.
Longer answers live in the docs FAQ.
Not specifically. Manual test management works before an LLM is connected. To use MCP, runs, and AI workflows, connect and validate any supported provider in workspace settings: a hosted API, a self-hosted model, or a supported sign-in provider.
Nothing. Suitest is Apache-2.0 licensed and self-hosted. There is no hosted service to subscribe to and no feature gate tied to payment.
Any MCP client. Claude Code, Cursor, and Codex are tested first-class: one npx command registers the server, and npx @suiflex/suitest-mcp init writes the config for your IDE automatically. Want the full local platform with the dashboard? npx @suiflex/suitest onboard boots everything and wires your IDE in one step.
Yes. The blackbox engine needs a URL and test credentials. It discovers routes, detects and performs login, builds an interaction graph, generates Playwright tests, and runs them, all without repo access.
Per-test video, screenshots, DOM snapshots, console logs, network HAR, and execution traces. Failures also produce an agent-readable failure bundle so your coding agent can fix the underlying bug.
It covers both roles: a test management UI with cases, suites, runs, and traceability, plus a deterministic runner that drives Playwright, HTTP APIs, and Postgres through MCP providers. You can adopt either half first.
Suitest sends LLM requests only to the provider explicitly configured for the workspace. Choose a self-hosted provider to keep inference on your infrastructure. Provider secrets stay encrypted on the Suitest server and are never sent to the MCP client.
A composite GitHub Action runs the suite, updates a single PR comment with results and failure excerpts, and fails the check when tests fail: exit code 0 for pass, 1 for test failures, 2 for infrastructure errors.
Your stack, your LLM, your data. Run the whole platform on a laptop or a cluster.
Suitest: 9/10 passed
Evidence and full failure detail in the run report.