Loading ToolBench
FastMCP Server exposing Google Antigravity (AGY) as an autonomous coding worker for AI architect agents (Claude Desktop, Cursor, Cline, Roo Code). Enables executing coding tasks, analytical discussions, git diff inspections, and automated test runner execution within specified project workspaces.
Server defines 8 tools with complete input schemas and descriptions. Naming follows verb_noun convention (agy_execute_task, agy_chat, agy_get_diff, agy_run_tests, agy_start_task, agy_job_status, agy_cancel_job, agy_list_jobs). Descriptions are detailed (150-300 chars) and explain WHAT, WHEN, and prerequisites. However, output schemas are not documented, LLMs cannot infer response structure. Error handling guidance is absent; tools do not indicate retryability or recovery paths. Parameter descriptions are present but lack format constraints (e.g., timeout_seconds has min/max but no guidance on typical values). No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels (WRITE, READ_ONLY, REVERSIBLE). Composition is sound, tools chain well (start_task → job_status → cancel_job) and accept natural identifiers (workspace_path as filesystem path, not opaque ID).
Stop a running background job and kill its process tree. Cancels a background task started with agy_start_task, terminating the AGY process and any child processes.
Submit an analytical, architectural, or planning query in read-only mode without modifying the workspace. Use this tool for code reviews, architectural discussions, design feedback, and planning queries where AGY analyzes the codebase but does not write or modify files.
Executes an autonomous multi-step coding task in a target workspace using Google Antigravity (AGY). Spawns a specialized AGY autonomous coding agent inside the specified repository or project directory. AGY investigates the codebase, writes new files, modifies existing code, refactors modules, executes build or setup commands, and returns structured telemetry and execution results. Architect Guidance: - Use this tool when you need AGY to implement concrete features, fix bugs, refactor code, or perform multi-file code modifications in a workspace. - For read-only planning or architectural queries without modifying files, set mode='plan' or use the dedicated `agy_chat` tool. - Always provide clear, self-contained prompts with explicit requirements, constraints, file paths, and acceptance criteria.
Inspect git status and unified diff of the workspace. Returns git status summary, file-by-file change statistics, and optionally the full unified diff of all modifications. Useful for reviewing what has changed before committing or understanding the scope of modifications.
Output schemas not documented. LLMs cannot infer response structure for any tool. Responses from agy_execute_task, agy_start_task, agy_job_status, and agy_list_jobs lack field definitions, forcing LLMs to guess at downstream chaining (e.g., what field contains the job_id returned by agy_start_task?).
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Tools are marked with Risk levels (WRITE, READ_ONLY, REVERSIBLE) in metadata but not exposed via MCP tool annotations. LLMs cannot determine safety/idempotency from the protocol.
Error handling lacks recovery guidance. No tool description indicates what to do on failure (retry, ask user, fatal). agy_execute_task and agy_start_task can fail due to invalid workspace_path or timeout, but descriptions do not guide LLM recovery.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 63 | <=2025-11-25 | v2 |
Collect a background job's result, optionally waiting a bounded number of seconds. Polls the status of a background task started with agy_start_task. Can wait for completion or return immediately with current status.
Recover job IDs and see what background tasks are still running. Lists all active and recent background jobs with their current status and metadata.
Execute the workspace test suite and return diagnostics. Runs automated tests (pytest, unittest, jest, etc.) in the target workspace and returns structured test results, failure details, and execution diagnostics.
Launch a background autonomous coding task and return a job ID immediately. Starts an AGY execution in the background, returning a job_id that can be used with agy_job_status to poll for results. Enables multi-minute tasks against clients with per-call timeouts.
Parameter constraints incomplete. timeout_seconds has min/max in schema but descriptions lack guidance on typical values or what happens on timeout. workspace_path requires 'existing directory' but no validation error message is documented.
Rigor parameter documentation is dense and assumes deep knowledge of the AI Deep Reasoning Toolkit. Description includes URLs and benchmark details that distract from the core parameter purpose. Should simplify to: 'Reasoning protocol: standard (default), deep (for multi-design tasks), or off (disabled).'