Testing

Testing

Tests run with Vitest (npm test → vitest run). Configuration: vitest.config.ts includes tests/**/*.test.ts with a 30-second timeout and .ts extension priority.

There is no tests/helpers/mock-server.test.ts — the helpers are support code consumed by the integration tests.

There is also a standalone regression suite at test/stream-log.test.ts (singular test/, not tests/). It drives .omp/stream-log.py as a Python subprocess and guards the log-formatter regressions that broke the OMP CI pipeline in issue #76. It is not picked up by the default npm test because vitest.config.ts only includes tests/**/*.test.ts; run it explicitly with npx vitest test/stream-log.test.ts if you change the OMP log formatter.

The suite covers:

• Canonical event flow (agent_start, turn_start, tool_execution_start/end, message_end, agent_end) formatting without crashing.
• tool_execution_end text fields that are null, numeric, lists, or dicts — all coerced safely instead of raising TypeError during string joins.
• tool_execution_start args payloads that are strings, null, lists, or dicts — still producing a tool invocation line so the CI log shows the call was attempted.
• Non-string text in message_end and agent_end events.
• Malformed JSON lines skipped without aborting the formatter.

Layout

tests/
├── helpers/
│   └── mock-server.ts        # fetch mock + MCP-over-HTTP test harness
└── integration/
    ├── config.test.ts        # resolveConfig resolution ladder
    ├── errors.test.ts        # status/network mappers
    ├── server.test.ts        # /health, initialize, tools/list, invalid session
    └── tools.test.ts         # all four tools: success + 401 paths

There are no unit tests per tool file and no stdio entrypoint tests — coverage is integration-level, exercising tools through the real HTTP transport.

The test harness — tests/helpers/mock-server.ts

Provides two layers of fakes:

mockChronovaApi()

Monkeypatches globalThis.fetch with a vi.fn that matches request URLs against registered handlers (string includes or RegExp). Supports:

• respond(pattern, response) — persistent handler.
• respondOnce(pattern, response) — one-shot (sets consumed = true).
• restore() — restores the original fetch.
• callCount() — number of intercepted fetch calls.

Responses are wrapped in a real Response with Content-Type: application/json plus any extra headers (used to test the 429 Retry-After path in errors.test.ts).

MCP-over-HTTP test server

startMcpTestServer(app) spins the Express app up on an ephemeral port and exposes:

• request(mcpReq) — POSTs a JSON-RPC request to /mcp with Accept: application/json, text/event-stream and tracks mcp-session-id. Parses SSE data: lines via parseSse.
• initSession(server) — sends initialize (protocolVersion 2025-03-26) then notifications/initialized.
• callTool(server, name, args) — sends tools/call and returns result, throwing on res.error.

This lets tests drive the server exactly as a real MCP client would, without spawning a process.

Test config

Integration tests use a fixed TEST_CONFIG:

{ apiKey: "test-api-key", apiUrl: "https://chronova.test/api/v1", port: 3001, configSource: "env" }

and call createApp(TEST_CONFIG) directly (bypassing resolveConfig), so tests are deterministic regardless of the host's ~/.chronova.cfg.

What the tests assert

• server.test.ts — /health returns { status: "ok", version: VERSION }; initialize returns serverInfo.name = "chronova-mcp" and version = VERSION; tools/list returns exactly 4 tools with the expected sorted names; every tool has annotations.readOnlyHint: true and an inputSchema.type = "object"; an unknown Mcp-Session-Id yields HTTP 400 with "Invalid or expired session ID".

  Note: server.test.ts imports VERSION from src/version.js, which reads package.json#version at import time. This keeps the test assertions in sync with the published package version automatically and avoids the previous drift caused by a hard-coded version string.

• stream-log.test.ts — .omp/stream-log.py exits 0 when fed the canonical OMP JSONL event flow and does not crash on malformed tool args, non-string tool text, malformed message/agent text, or invalid JSON lines.

• tools.test.ts — for each tool: a happy path asserting parsed JSON content, a 401 path asserting isError: true and the "Unauthorized" message; plus parameter-passthrough checks (e.g. get_productivity_summary with project, get_recent_activity with filters/pagination).

• config.test.ts — resolveConfig priority: env wins over ~/.chronova.cfg, which wins over ~/.wakatime.cfg, which wins over none; uses injected readFile/getHomeDir/env so no real filesystem access.

• errors.test.ts — mapHttpStatusToError for 401/404/429/5xx/generic; 429 retryAfter from Retry-After and from X-RateLimit-Reset; mapNetworkError produces CONNECTION_ERROR.

Running

npm test                        # vitest run (CI mode, tests/ only)
npx vitest                      # watch mode
npx vitest test/stream-log.test.ts  # OMP log-formatter regression suite
npm run type-check              # tsc --noEmit, no tests

No test runner script is needed beyond vitest run; there is no separate e2e suite or coverage threshold configured. The stream-log.test.ts path is in the separate test/ directory, not tests/integration/, and must be invoked explicitly.