Loading ToolBench
A local MCP bridge for delegating tasks to Google Antigravity CLI
Server defines 5 tools with explicit schemas and descriptions. All tools have input schemas with Zod validation and descriptions present. However, descriptions are verbose policy statements rather than concise LLM-optimized guidance (baseline 50-200 chars; these are 300-800+ chars). Output schemas are not documented. Error handling exists but lacks recovery guidance. Tool naming follows verb_noun pattern (antigravity_run, antigravity_continue, antigravity_fetch, antigravity_batch, antigravity_models) but lacks clarity on distinctions. Parameters are well-typed with enums and constraints, but descriptions embed policy text rather than actionable parameter guidance.
Execute a directed acyclic graph (DAG) of isolated tasks and verification gates with bounded retry, producing structured digests and artifacts.
Policy: Host acts only as orchestrator to set direction, allocate nonoverlapping scopes, and review final diffs and evidence. Delegate ALL repository investigation, web search, source fetching, implementation, tests, and corrections to agy; host must not perform broad file investigation, web search, or code editing directly. Provide a concise prompt with explicit file ownership boundaries and test criteria, asking for a compact handoff of changed files, verification commands and results, fetched source URLs and facts, and any blockers. The delegated agy worker executes tasks directly using native tools without recursively invoking agy-orch-mcp or delegating back. Follow up in an existing Antigravity conversation. Different explicit conversation_ids can run in parallel; simultaneous continuations of the same ID return BUSY. Without an ID, agy resumes its most recent conversation and requires exclusive access to this server, otherwise BUSY is returned. Other CLI sessions may change the latest conversation. Use the same workspace as the original turn.
Fetch a selective slice of an artifact from a prior run by artifact_id or logical selector (e.g. task:<id>:stdout, task:<id>:stderr, gate:<id>:stdout, manifest, digest). Never accepts raw file paths.
List available model slugs and display names using agy models. Does not start a model turn; may contact the Antigravity service.
Tool descriptions are policy statements (300-800 chars) rather than concise LLM-optimized guidance. Baseline is 50-200 chars. Descriptions should state WHAT the tool does, WHEN to use it, and key prerequisites, not embed full policy text.
Output schemas are not documented. LLMs cannot plan downstream calls or extract required fields without knowing what fields are returned. All tools lack return type documentation.
Error handling returns status codes and error messages but lacks recovery guidance. Errors should tell the LLM what to do next (retry, ask user, or fatal). E.g., 'CONFIG_ERROR' gives no actionable path.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | 2025-06-18+ | v2 |
Policy: Host acts only as orchestrator to set direction, allocate nonoverlapping scopes, and review final diffs and evidence. Delegate ALL repository investigation, web search, source fetching, implementation, tests, and corrections to agy; host must not perform broad file investigation, web search, or code editing directly. Provide a concise prompt with explicit file ownership boundaries and test criteria, asking for a compact handoff of changed files, verification commands and results, fetched source URLs and facts, and any blockers. The delegated agy worker executes tasks directly using native tools without recursively invoking agy-orch-mcp or delegating back. Start one Antigravity CLI turn in a new conversation. Returns a conversation_id for follow-up. Up to 4 agy calls can run concurrently per server; excess calls return BUSY. Parallel calls share workspace files, Antigravity account and quota.
Tool naming lacks distinction clarity. 'antigravity_run' vs 'antigravity_continue' vs 'antigravity_batch' are not obviously different to an LLM. Names should make the distinction obvious (e.g., 'antigravity_start_task', 'antigravity_resume_task', 'antigravity_execute_dag').
Parameter descriptions embed policy constraints rather than actionable guidance. E.g., 'autonomy' param description is 150+ chars of policy text. Should be: 'Permission level: safe (inherit agy settings), sandbox (restrict terminal), full (bypass permissions, requires opt-in).'