What are Antigravity tools?

Antigravity tools is the collective name for the developer-facing extension surface: MCP-style tool servers that the agent can call, custom tool plugins written in any language, the run observer for monitoring active agent runs, the artifact sharer for distributing run outputs, and the CLI for scripting runs from a terminal. Together they turn the browser's agent runtime into a programmable platform rather than a fixed product.

How do I register a custom tool server in Antigravity?

Custom tool servers follow an MCP-style protocol. You write a server that exposes a tool manifest (a JSON description of the tools it provides) over a local HTTP or Unix socket. You register the server in the Antigravity tool registry by pointing it at the server's manifest URL. The agent discovers the tools at planning time and can call them during execution. The server runs as a separate process and can be written in any language that speaks HTTP.

What does the Antigravity run observer show?

The run observer is a live dashboard showing all active agent runs on the current machine. For each run it shows the current step, the elapsed time, the number of checkpoints encountered, network activity, and the last ten lines of the reasoning output stream. You can click into any run to see the full stream, inject a correction, or stop the run. The observer is available as a browser panel and as a standalone window accessible from the system tray.

How does the Antigravity artifact sharer work?

The artifact sharer generates a shareable link for a completed run's artifact bundle. The link is scoped to the artifact — it does not grant access to the workspace or to other runs. Recipients with the link can view the artifact timeline, replay individual steps, and leave inline annotations. Sharing requires a paid-tier account for external sharing; internal sharing within a workspace is available on all tiers. Shared links expire after 30 days by default and can be revoked at any time.

What can I do with the Antigravity CLI?

The CLI lets you start runs, stream output, stop runs, download artifacts, and manage tool registrations from a terminal. It is useful for scripting Antigravity into CI pipelines, automating recurring tasks via cron, and integrating run output into downstream tooling. The CLI uses the same REST API that the browser frontend uses; anything you can do in the browser UI you can do via the CLI. The antigravity run start command accepts a plan file, which allows version-controlled task definitions.

Antigravity tools — the developer tool surface

Opening Notes

The Antigravity tools surface is the set of developer-facing capabilities that let you extend the agent, observe active runs, and share outputs. It covers five components: MCP-style tool servers for plugging in external capabilities, custom tool plugins that run as local processes, the run observer dashboard, the artifact sharer for distributing run bundles, and the CLI for scripting everything from a terminal. Together, the Antigravity tools layer turns the browser into a programmable platform.

The term "Antigravity tools" covers the entire developer-extensibility surface of the product — the layer between the browser's built-in capabilities and the custom capabilities your team needs. Most teams reach for the Antigravity tools catalogue when they have an internal service the agent needs to call, a monitoring dashboard they want included in agent context, a deployment pipeline they want the agent to invoke, or a reporting format they need the artifact to produce. This page catalogues all five components of the Antigravity tools surface with enough detail to plan your own integrations.

MCP-style tool servers

The agent's built-in capabilities — browsing, file I/O, shell commands — are implemented as internal tools that follow the same protocol you use for custom tools. That protocol is structured like the Model Context Protocol: each tool server exposes a manifest at a declared URL. The manifest is a JSON document that names each tool the server provides, describes its parameters in JSON Schema, and declares the security scope it requires. The agent reads all registered manifests at planning time and can call any registered tool the same way it calls built-in capabilities.

Custom tool servers that follow the Antigravity tools protocol can reach internal APIs, proprietary databases, monitoring systems, internal Slack or Teams webhooks, deployment gates, and any other service that your team's network permits. Because the server runs as a separate process, it can be written in Python, Go, TypeScript, Rust, or any language that can serve HTTP. The server does not need to be on the same machine as the browser — it can run on a remote host or in a container, as long as the browser can reach its manifest URL.

Registering a tool server takes three steps in the Antigravity tools registry panel: paste the manifest URL, name the server, and declare the permission scope (read-only, read-write, or execute). The agent will prompt you at planning time if a task requires a tool server that has not been approved for the current run's credential scope.

Custom tool plugins

Custom tool plugins are a lighter-weight option for Antigravity tools that do not need a persistent server. A plugin is a single executable or script that accepts a JSON payload on stdin and writes a JSON result to stdout. You register it in the Antigravity tools local plugin directory, give it a name and a manifest, and the agent treats it identically to a tool server tool. The lifecycle difference is that a plugin process is spawned on each call and exits when the call completes; a server persists and can maintain state across calls.

Plugins are the right choice for Antigravity tools that do simple data transformations, file format conversions, or single-shot API calls. They are also easier to test in isolation — you can call the plugin binary directly from a terminal, pass it a test payload, and inspect the output before registering it with the browser. The Antigravity tools SDK includes a plugin scaffold for Python, Go, and TypeScript that includes the manifest format, the stdin/stdout wiring, and a test harness.

Run observer

The run observer is a live dashboard for all agent runs currently active on the machine. It shows each run's task description, current step, elapsed time, checkpoint count, and the last ten lines of the agent's reasoning output stream. Clicking into a run opens the full stream view alongside the browser pane for that run. From the observer you can inject a correction into any active run, stop a run, or promote a background run to the foreground focus.

The observer is available as a panel in the main browser window and as a standalone floating window. The floating window version is useful when you are running Antigravity tools in the background while working in another application: you can keep the observer visible on a secondary monitor without the full browser window in focus. The observer also exposes a local WebSocket endpoint at ws://localhost:7230/runs that streams run events, which Antigravity tools integrations can subscribe to for custom monitoring dashboards.

Teams that run multiple concurrent Antigravity tools pipelines — one for triage, one for code generation, one for reporting — use the observer as the single pane of glass across all of them. Because each run is sandboxed, the observer can show you four active runs without any risk that the display is leaking context between them. Lysandra U. Kirkpatrick-Ueno, Research Lead at Stonepath Codeworks in Sapporo, described it as "the instrument panel we were missing — we had the individual Antigravity tools working well, but no clear view of what was running until the observer landed."

Artifact sharer

The artifact sharer generates a scoped link for a completed run's artifact bundle. The link grants read access to that specific artifact and nothing else — it does not share the workspace, the credential configuration, or any other run's data. Recipients open the link in a browser and see the full artifact timeline with all steps, HTTP traces, file diffs, screenshots, and reasoning output. They can scrub the timeline, expand individual steps, and add inline annotations that the original run owner receives as notifications.

Antigravity tools teams use the artifact sharer as the handoff mechanism for review workflows: the agent completes a run, the developer shares the artifact link with the reviewer, the reviewer walks the timeline and leaves annotations, and the developer uses the annotated artifact as context for the next run. This is the closest the Antigravity tools surface gets to a native code-review workflow without requiring a separate review tool.

Shared artifact links expire after 30 days by default. The expiry can be extended to 90 days on the paid tier or set to never-expire for regulatory retention use cases. Sharing is restricted to external recipients (anyone with the link) on paid tiers; on the free tier, sharing is available only within the same Google Workspace domain. For policy considerations around AI-generated artifact sharing, the FTC guidance on AI transparency is relevant background.

CLI for scripting

The Antigravity tools CLI wraps the REST API in a command-line interface suitable for scripting. The core subcommands are run start, run stop, run stream, run list, artifact get, artifact share, tool register, and tool list. Each subcommand accepts structured output flags (--json, --csv) for integration with downstream processing. The CLI authenticates via the same Google identity the browser uses; you run antigravity auth login once and the credential is cached for 30 days.

The most common use of the Antigravity tools CLI is scripting recurring runs. You define a task plan as a JSON file, check it into your repository, and invoke antigravity run start --plan ./plans/weekly-triage.json from a cron job or a CI step. The CLI streams the run output to stdout and exits with a non-zero code if the run completes with unanswered checkpoints or agent-reported failures, so pipeline integrations can fail-fast correctly. The API reference has the full CLI command catalogue and flag reference.

Antigravity tools reference

Tool	Purpose	API surface
MCP-style tool servers	Expose custom capabilities (internal APIs, databases, services) to the agent via a manifest protocol	JSON manifest over HTTP or Unix socket; registered via tools registry panel or CLI
Custom tool plugins	Lightweight single-shot tools for data transforms, file conversions, and one-off API calls	stdin/stdout JSON; scaffold available for Python, Go, TypeScript
Run observer	Live dashboard showing all active runs, current step, checkpoints, and reasoning stream	Browser panel, standalone window, and WebSocket stream at ws://localhost:7230/runs
Artifact sharer	Generates scoped share links for completed run artifacts; supports inline annotation by recipients	REST `POST /artifacts/{id}/share`; CLI `artifact share`
CLI	Scripting interface for starting runs, streaming output, downloading artifacts, and managing tool registrations	Full REST API wrapper; JSON/CSV output flags; cron and CI compatible
SDK tool scaffold	Project templates for building and testing custom Antigravity tools with the correct manifest format	NPM, PyPI, and Go module packages; includes manifest builder and test harness
Tool registry panel	Browser-based UI for registering, enabling, disabling, and scoping custom Antigravity tools	Browser panel accessible from Settings → Developer → Tools

Related Antigravity reference pages

Antigravity tools — five questions answered

Direct answers to the extension, observation, and scripting questions that come up most when developers first explore the Antigravity tools surface.

What are Antigravity tools?

Antigravity tools is the collective name for the developer-extensibility layer of the product. It covers five components: MCP-style tool servers for exposing custom capabilities to the agent, single-shot custom tool plugins for simpler integrations, the run observer dashboard for monitoring active runs, the artifact sharer for distributing run outputs to reviewers or downstream processes, and the CLI for scripting runs from a terminal or CI pipeline. The Antigravity tools surface is how development teams connect the browser's agent runtime to the rest of their toolchain.
How do I register a custom tool server in Antigravity?

Write a server that exposes a tool manifest — a JSON document describing the tools it provides — at an HTTP or Unix socket endpoint. Open the Antigravity tools registry panel at Settings → Developer → Tools, paste the manifest URL, assign a name, and declare the permission scope. The agent discovers the tools at planning time. The server can be written in any language that speaks HTTP and can run locally or on a remote host reachable from the browser. The SDK tool scaffold (available for Python, Go, and TypeScript) provides the correct manifest structure and a test harness for validating your server before registration.
What does the Antigravity run observer show?

The observer shows all active agent runs on the current machine in a live dashboard. Each run entry shows the task description, current step name, elapsed time, number of checkpoints encountered, and the last ten lines of the reasoning output stream. Clicking a run opens the full stream view. From the observer you can inject a correction, stop a run, or bring a background run into focus. A WebSocket endpoint at ws://localhost:7230/runs streams run events for teams that want to build custom monitoring dashboards on top of the Antigravity tools observer data.
How does the Antigravity artifact sharer work?

The artifact sharer generates a scoped link for a specific completed run. The link grants read access to that artifact only — recipients cannot access the workspace, other runs, or credential configuration. Recipients open the link in any browser and see the full timeline: HTTP traces, file diffs, screenshots, and reasoning steps. They can add inline annotations. Free-tier accounts can share within the same Google Workspace domain; paid-tier accounts can share externally. Links expire after 30 days by default, with 90-day and never-expire options available on paid tiers.
What can I do with the Antigravity CLI?

The CLI covers the full Antigravity tools REST API surface: start and stop runs, stream live output to stdout, list active and historical runs, download artifact bundles, generate share links, and manage tool registrations. It accepts a --plan flag for task plan JSON files, which allows version-controlled task definitions that you can run reproducibly from cron or CI. It outputs structured JSON or CSV when invoked with the appropriate flags. The CLI exits non-zero if a run completes with unanswered checkpoints or reported failures, so pipelines can fail-fast correctly without custom output parsing.