A thin, auditable MCP server wrapping the Antigravity CLI (agy). Forward-compatible with future CLI flags via extra_args passthrough.
Single tool 'agy' has a comprehensive input schema with 11 parameters, all properly typed and described. Tool description is clear and actionable (127 chars). Parameters cover a wide range of use cases (query, cwd, add_dir, sandbox, conversation, timeout, env_overrides, stdin_text, etc.) with good granularity. However, output schema is not documented, the tool returns 'structured JSON output from agy's --output-format json envelope' but the actual response structure (fields, types) is not specified. This forces LLMs to guess at response shape and limits downstream tool chaining. Error handling is present in code (directory validation, launch failures, timeouts) but not surfaced in the tool description. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite the tool having WRITE risk. Parameter descriptions are generally good (60-100 chars each) but some lack format/constraint details (e.g., 'timeout' lacks min/max bounds).
Execute the Antigravity CLI (agy) with a query and optional parameters. Returns structured JSON output from agy's --output-format json envelope.
Output schema not documented. Tool description states 'Returns structured JSON output from agy's --output-format json envelope' but does not specify the response structure (field names, types, required fields). LLMs cannot plan downstream operations or extract specific fields without knowing the response shape.
Tool annotations missing. Tool has WRITE risk (executes external CLI, modifies state via --sandbox, --continue flags) but lacks destructiveHint or idempotentHint annotations. LLMs cannot reason about retry safety or side effects.
Numeric parameter 'timeout' lacks bounds. Description says 'Hard timeout in seconds for agy execution (default 660)' but does not specify min/max values. LLMs may pass absurd values (0, 999999) that break execution.
Error handling not surfaced in tool description. Code handles directory validation, launch failures, and timeouts, but the tool description does not explain what errors are possible or how to recover. LLMs cannot anticipate failures.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | <=2025-11-25 | v2 |
Parameter 'print_timeout' lacks format specification. Description says 'Timeout value to pass to --print-timeout flag' but does not clarify if it accepts seconds, milliseconds, or a duration string (e.g., '30s'). LLMs may pass invalid formats.