- name
- mcp-testing
- description
- MCP testing guidance and patterns. Loaded by /ops/test/mcp command or subagents for comprehensive Navigator MCP server testing.
- compatibility
- Requires bun, jq, and navigator-server running on :9334
- allowed-tools
- Read Glob Grep Skill TodoWrite Bash(./.claude/scripts/run-mcp-tests.sh *) Bash(nav *) Bash(nav-dev *)
- metadata
- {"author":"outfitter-dev","version":"1.0"}
# Navigator MCP Testing
Automated tests for Navigator MCP server via JSON-RPC protocol.
## Quick Start
```bash
# Start server
bun run dev
# Run tests
./.claude/scripts/run-mcp-tests.sh schema-validation
./.claude/scripts/run-mcp-tests.sh --all
```
## Structure
```
.claude/
├── commands/ops/test/mcp.md # Command definition
├── scripts/run-mcp-tests.sh # Test runner (all logic here)
├── scripts/lib/test-runner-lib.sh # Shared test library
└── skills/mcp-testing/SKILL.md # This file
```
## Test Categories
| Category | Tests | What It Validates |
|----------|-------|-------------------|
| `schema-validation` | 7 | Zod schema validation, missing/invalid params |
| `action-routing` | 7 | Action dispatch for all action types |
| `error-responses` | 4 | Error codes and structured error handling |
| `response-formatting` | 4 | MCP content structure (text/image types) |
## Output
```
.scratch/testing/
├── 20260116-abc12-mcp-schema-validation.md # Results table
├── 20260116-abc12-mcp-schema-validation-debug.log # Debug output
└── ...
```
## Adding Tests
Edit `.claude/scripts/run-mcp-tests.sh`:
```bash
# Add a test to existing category
run_mcp_test 8 "New test name" '{"action":"snap"}' "expected_pattern" "false" # false = expect success
# Add new category
run_new_category() {
setup_category "new-category"
log "Running new-category tests..."
# Setup: navigate to test page
mcp_call '{"action":"navigate","url":"https://example.com"}' >/dev/null 2>&1
run_mcp_test 1 "Test name" '{"action":"..."}' "pattern" "false"
# ... more tests
finalize_category
}
```
Then add to `main()`:
```bash
new-category) test_new_category || exit_code=1 ;;
```
## Key Differences from CLI Tests
| Aspect | CLI | MCP |
|--------|-----|-----|
| Protocol | HTTP to server | JSON-RPC over stdio |
| Input | Shell arguments | JSON objects |
| Output | JSON stdout | MCP content array |
| Error detection | Exit code + stderr | `isError` flag |
## Exit Codes
- `0` - All passed
- `1` - Failures
- `2` - Setup error (server not running, jq missing)
Ver no GitHub