Swipium

MCP server for simulator-based mobile QA agents

QA Tools for Your Agent

Test and catch issues before the build.

Swipium is free and open source. It exposes simulator-based mobile QA as structured MCP tools so agents can launch apps, inspect screens, execute UI actions, collect evidence, build app maps, generate reports, and create reusable test assets.

GitHub Follow Swipium

What it is

An agent-facing QA harness.

Swipium runs locally as a stdio MCP server. Agents communicate with it through structured JSON-RPC tools for simulator setup, observation, UI actions, smoke checks, exploration, evidence capture, reporting, app-map memory, flow generation, suite generation, and automation output.

The goal is deterministic feedback from a usage perspective. Teams can detect product and flow issues before release, with artifacts that make each agent run auditable, inspectable, and repeatable.

Core capabilities

Built around simulator QA execution loops.

Simulator control

Prepare Android Emulator or iOS Simulator targets, launch apps, capture snapshots, and execute UI actions.

Evidence first

Collect screenshots, logs, dumps, and reports that explain what the agent observed, and export them for CI as JUnit or SARIF.

Durable app map

Persist screens, features, flows, coverage context, issues, and generated test cases under the tested project.

Reusable outputs

Convert exploration into flows, test cases, suites, and generated automation for future runs.

Integrations

Connect Swipium to the agent environment you already use.

Swipium works through MCP, so it can support all models available in the connected agent client. Use it with Claude Code, Codex, Gemini CLI, Cursor, VS Code, Grok, Grok Build, or any MCP-compatible workflow that can launch a local server.

Claude Code

Use Claude Code with Fable 5.1, Opus 5.5, Sonnet 5.5, and any model exposed by the client. Project-scoped setup keeps QA tools tied to the mobile repository. Or install the plugin: /plugin marketplace add GeroPalombo/swipium.

npm install -g swipium
swipium init claude \
  --scope project --apply

Gemini CLI

Connect Gemini CLI to Swipium so simulator checks, evidence capture, and reports run locally through MCP.

npm install -g swipium
swipium init gemini --apply

Codex

Use Codex with GPT-6.1 Sol and any model available in the client. Run swipium init codex so Codex passes test credentials and the Android SDK path to Swipium, then inspect mobile apps, drive UI flows, and create reusable QA assets from the project root.

npm install -g swipium
swipium init codex --apply

Cursor and VS Code

Register Swipium in the editor you already use. Project-scoped setup writes a portable config the team can commit.

npx -y swipium init cursor \
  --apply
npx -y swipium init vscode \
  --apply

Grok and Grok Build

Use Grok or Grok Build through MCP-compatible clients or bridges to route simulator QA tasks into Swipium.

npx -y swipium verify
npx -y swipium
View integration details

Public 2.0 scope

Local, simulator-first, explicit where risk exists.

Documentation

Start with the autopilot tool, then inspect the evidence.

Open docs on GitHub

Register and verify

npx -y swipium init claude \
  --scope project --apply
npx -y swipium verify

Configure an MCP client

{
  "mcpServers": {
    "swipium": {
      "command": "npx",
      "args": ["-y", "swipium"],
      "env": {
        "SWIPIUM_PROJECT_ROOT":
          "/path/to/app"
      }
    }
  }
}

Ask an agent

Smoke test this app with
Swipium: qa_test_this with
goal "smoke" and
waitForCompletion. Then
read report.md.

Blog

Swipium 2.3.0 new release banner

October 12, 2026

Swipium 2.3.0

Faster, more precise and cheaper in tokens: three calls to a tested app, taps that land on the right thing, half-size results, and verdicts that are harder to fool.

Read the release notes