@kaneshir/lisa-mcp
v1.0.2
Published
Lisa MCP server — precompiled binary for AI-driven browser automation and QA intelligence
Maintainers
Readme
@kaneshir/lisa-mcp
The Lisa MCP server as a precompiled npm package. No Dart SDK required.
Lisa is an AI QA intelligence layer with 24 MCP tools for browser automation, test recording, flow management, persona generation, and flakiness tracking. This package ships platform-native binaries so any Node.js project can spawn the server without a Dart installation.
Install
npm install @kaneshir/lisa-mcpVerify the binary is working:
npx lisa-mcp verifyExpected output:
lisa-mcp verify
binary: /path/to/node_modules/@kaneshir/lisa-mcp/bin/lisa_mcp-macos-arm64
status: executable ✓
version: 1.0.2
mcp handshake: ✓Platform support
| Platform | Binary |
|----------|--------|
| macOS Apple Silicon (M1+) | lisa_mcp-macos-arm64 |
| macOS Intel | lisa_mcp-macos-x64 |
| Linux x64 | lisa_mcp-linux-x64 |
| Windows | Not yet supported |
Usage with Synthetic Test Fabric
The primary consumer is synthetic-test-fabric. In your BrowserAdapter.generateFlows() implementation:
import { spawn } from 'child_process';
import path from 'path';
import { buildLisaMcpCommand } from '@kaneshir/lisa-mcp';
async function generateFlows(iterRoot: string, candidates: CandidateFlow[]) {
const { cmd, args } = buildLisaMcpCommand();
const server = spawn(cmd, args, {
env: {
...process.env,
LISA_MEMORY_DIR: path.join(iterRoot, '.lisa_memory'),
},
stdio: ['pipe', 'pipe', 'inherit'],
});
// Send MCP tool calls over server.stdin (JSON-RPC)
// Read tool responses from server.stdout
// Use lisa_explore, lisa_get_seeded, lisa_codegen to generate Playwright specs
}See synthetic-test-fabric docs/lisa-mcp.md for the complete integration guide.
Usage with Claude Code
Lisa's 24 MCP tools are also available directly to Claude Code sessions. Get the config:
npx lisa-mcp config /path/to/.lisa_memoryOutput:
{
"mcpServers": {
"lisa": {
"command": "/path/to/lisa_mcp-macos-arm64",
"args": [],
"env": {
"LISA_MEMORY_DIR": "/path/to/.lisa_memory"
}
}
}
}Add this to your Claude Code MCP settings.
MCP tools
24 tools exposed via MCP stdio JSON-RPC:
| Tool | Purpose |
|------|---------|
| lisa_health | Server status, memory usage, metrics |
| lisa_keys | Widget key management (store / get / stats) |
| lisa_batch | Execute multiple tools in parallel |
| lisa_explore | Parse element keys, detect screen state, suggest actions |
| lisa_setup | Load flows, import seeds, start session |
| lisa_analyze | Diagnostics, cache stats, key stats |
| lisa_report | Coverage, bug history, flaky elements, priorities |
| lisa_action | Chrome command builder (click / type) |
| lisa_failure | Test failure triage (classify / heal / report) |
| lisa_flow | Flow management (save / list / progress) |
| lisa_run_flow | Flow execution lifecycle (start / complete / check) |
| lisa_codegen | Test code generation (generate / validate / flow_to_test) |
| lisa_orchestrate | Master workflow orchestration |
| lisa_record | Recording sessions (start / stop / event / export) |
| lisa_api_setup | API testing setup |
| lisa_api_step | API flow execution (start / next / report / complete) |
| lisa_get_seeded | Query seeded test entities from lisa.db |
| lisa_route_decision | Intelligent decision routing with LLM fallback |
| lisa_prepare_for_action | Handle dialogs and loading states |
| lisa_flakiness | Flakiness tracking (index / report / drive) |
| lisa_create_flow | Natural language → structured regression flow |
| lisa_generate_personas | Generate persona YAMLs from behavior cache |
| lisa_form_hypothesis | Record a testable hypothesis before a test run |
| lisa_record_hypothesis_outcome | Log the outcome of a previously formed hypothesis |
Programmatic API
import {
buildLisaMcpCommand, // { cmd, args } for spawning
buildLisaMcpConfig, // Claude Code MCP server config entry
resolveBinaryPath, // absolute path to the binary
verifyBinary, // async health check
} from '@kaneshir/lisa-mcp';
// Spawn the server
const { cmd, args } = buildLisaMcpCommand();
// Get Claude Code config
const config = buildLisaMcpConfig({ memoryDir: '/path/to/.lisa_memory' });
// Verify
const result = await verifyBinary();
console.log(result.ok, result.version);Troubleshooting
permission denied
chmod +x node_modules/@kaneshir/lisa-mcp/bin/lisa_mcp-*unsupported platform 'win32-x64'
Windows support is on the roadmap. Track progress at the dev-infra issue tracker.
Binary not found after install
rm -rf node_modules/@kaneshir/lisa-mcp && npm install @kaneshir/lisa-mcpMCP handshake fails
Check LISA_MEMORY_DIR is a valid writable directory. The server creates lisa.db there on first start.
License
MIT — source in private kaneshir/dev-infra repository. Only precompiled binaries are distributed via this package.
