Antigravity MCP Server

Overview Schema Source Related Servers Discussions Admin

delegate-antigravity

Destructive

Delegates long, multi-step implementation tasks to Antigravity in an isolated git worktree, returning a task ID for progress polling, follow-up instructions, and final diff retrieval.

Instructions

Delegate a long, multi-step implementation task to Antigravity. Runs detached in its own git worktree/branch, by default on the Orchestrator model (planner + Gemini 4 Hydrogen implementer, built for long tasks). Returns a taskId immediately; poll status for live progress, continue to send follow-up instructions into the same conversation, and result for the final answer + diff. The worktree isolates git state only — with skipPermissions (default) the agent can run any command.

Tool call template

This tool declares no required parameters. Add optional parameters from the input schema as needed.

{
  "id": 1,
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "arguments": {},
    "name": "delegate-antigravity"
  }
}

Input Schema

Name	Required	Description	Default
`mode`	No	Agent mode (--mode): accept-edits, auto (sandboxed autonomy; not available on every account), or plan (answers go into a plan artifact).
`task`	No	What Antigravity should accomplish (required for start).
`model`	No	Model ID or label. Omit to use the Orchestrator (Carbon, else Argon) when your plan has one, otherwise Gemini 4 Hydrogen (High).
`action`	No	start: spawn a delegated task; status: live progress; continue: send a follow-up turn to the same conversation; result (alias collect): final answer + diff; stop: kill the running turn; list: all delegations.	start
`effort`	No	--effort for non-Orchestrator models (Orchestrator has a fixed thinking level).
`signal`	No	Signal for action=stop (default SIGTERM; SIGKILL for a run that ignores it).	SIGTERM
`taskId`	No	Delegation ID returned by start (any turn's task ID also works).
`cleanup`	No	On result/collect: also remove the worktree and delete the branch after grabbing the diff.
`message`	No	Follow-up instruction for action=continue.
`sandbox`	No	--sandbox: run terminal commands in the OS sandbox.
`timeout`	No	Per-turn --print-timeout in milliseconds (default 7200000 = 2h; 0 = no limit).
`autoTest`	No	Command the agent must run to verify its work (e.g. "npm test"); it is told to loop until it passes.
`baseBranch`	No	Branch/commit to base the worktree on (default: current HEAD of repo).
`branchName`	No	Name for the new branch in the worktree. Default: agy/<random>.
`workingDir`	No	Repo root (default: resolved automatically).
`skipPermissions`	No	--dangerously-skip-permissions (default true: a headless run cannot answer permission prompts; false when `mode` is set; not allowed with mode plan). The worktree isolates git state, NOT the filesystem — the agent can still read/write outside it and run any command.
`acceptanceCriteria`	No	Bullet list of criteria the implementation must satisfy.

Schema Changelog

Changes observed during successful MCP inspections.

First observedv0.3.0Nov 14, 2026

TDQS

A4.3/5.0

Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Adds substantial context beyond the annotations: detached execution in its own git worktree/branch, immediate taskId return, and the critical caveat that the worktree isolates git state only — with skipPermissions defaulting true the agent can run any command and touch files outside the worktree. That directly reinforces the destructive/openWorld hints with concrete consequences. It could go further on what stop/cleanup actually destroy, but the safety picture is well disclosed.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Four dense sentences, front-loaded with purpose then lifecycle then the safety caveat. Every sentence carries information, though the parenthetical model-stack detail is slightly heavy for the space it occupies.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a 17-parameter, no-output-schema tool this covers the essentials: what it does, how to poll, what result returns, and the isolation caveat. Minor gaps remain around cleanup semantics and what stop does to in-flight work, but nothing an agent needs to invoke it correctly is missing.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 100%, so the baseline is 3; the description earns above that by explaining the action lifecycle and defaults (Orchestrator model selection, skipPermissions default true, 2h timeout) in prose that ties parameters to behavior. It does not add syntax detail beyond the schema, so it is not a 5.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

States a specific verb and resource ('Delegate a long, multi-step implementation task to Antigravity') and immediately distinguishes itself from siblings by naming the follow-up tools (status, continue, result) and the detached worktree execution model. An agent can tell this apart from ask-antigravity or plan-antigravity without opening the schema.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Explicitly scopes usage to 'long, multi-step implementation' tasks and lays out the polling lifecycle (status for progress, continue for follow-ups, result for final answer + diff). It does not explicitly state when to prefer a sibling like ask-antigravity for short queries, so it stops short of full when-not guidance.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Deploy Server