agy-mcp-server
Provides tools for interacting with Google Gemini models via the Antigravity CLI (agy), enabling planning, code review, test validation, and general prompts with a selection of Gemini models.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@agy-mcp-server create an implementation plan for adding PostgreSQL caching"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
agy-mcp-server
Languages: English | Português
An MCP server that exposes the Antigravity CLI (agy) to MCP clients such as OpenCode. It provides structured tools for planning, code review, test validation, generic prompts, and literal CLI commands.
The server is intentionally restricted to Google Gemini models and defaults to gemini-3.7-flash-medium.
Important: This project requires the official
agyCLI to be installed and authenticated. It does not provide or redistribute Antigravity credentials or the CLI itself.
Features
MCP server over stdio for local clients
agy_planfor architecture and implementation plansagy_reviewfor technical and security-oriented code reviewagy_validate_testsfor test quality and coverage analysisagy_execfor general-purpose Antigravity promptsagy_commandfor literal slash commands such as/usageStrict allowlist for Google Gemini models
Safe argument passing through
spawn, without shell interpolationConfigurable project context through
directoryFive-minute execution timeout with graceful termination and force-kill fallback
English tool descriptions and error messages
Related MCP server: antigravity-mcp
Requirements
Node.js 20 or later
npm
The
agyCLI installed and authenticatedAn MCP-compatible client
By default, the server looks for agy at ~/.local/bin/agy. Set AGY_BIN_PATH when it is installed elsewhere.
Installation
git clone https://github.com/moisesfilho/agy-mcp-server.git
cd agy-mcp-server
npm ci
npm run buildOpenCode Configuration
Add the following entry to ~/.config/opencode/opencode.json:
{
"mcp": {
"agy-proxy": {
"type": "local",
"command": [
"node",
"/absolute/path/to/agy-mcp-server/dist/index.js"
]
}
}
}For a local installation in the standard project directory, use:
{
"mcp": {
"agy-proxy": {
"type": "local",
"command": [
"node",
"/home/moises/Projetos/agy-mcp-server/dist/index.js"
]
}
}
}Restart the MCP client after changing its configuration.
Available Models
Every tool accepts an optional model argument. Only the following models are accepted:
Model | Default |
| Yes |
| No |
| No |
| No |
| No |
| No |
| No |
| No |
| No |
| No |
| No |
The model suffix controls the reasoning effort. The server does not send a separate --effort flag because the agy CLI rejects conflicting combinations.
Usage
Ask the MCP client to use a tool. Examples:
Create a plan
Use agy_plan to create an implementation plan for adding PostgreSQL caching to this project.Tool arguments:
{
"prompt": "Create an implementation plan for adding PostgreSQL caching.",
"directory": "/home/moises/Projetos/my-project",
"model": "gemini-3.7-flash-medium"
}Review code
Use agy_review to inspect the current changes for security issues and regressions.Validate tests
Use agy_validate_tests to identify missing edge cases in the current test suite.Run a direct CLI command
Use agy_command when the input must reach agy literally, without an LLM interpreting it:
{
"command": "/usage"
}This is equivalent to:
agy -p /usage --output-format textOther slash commands can be sent in the same way, provided they are supported by the installed agy version.
Run a general prompt
{
"prompt": "Analyze the current architecture and identify the three largest reliability risks.",
"mode": "plan",
"directory": "/home/moises/Projetos/my-project",
"model": "gemini-3.7-flash-high"
}Configuration
Variable | Default | Description |
|
| Absolute path to the |
The server does not store credentials. Authentication is handled by the local agy installation and its own configuration.
Security Notes
The server passes arguments directly to the child process and does not invoke a shell.
agyis started with--dangerously-skip-permissions, as required for unattended MCP calls. Review prompts and workspace permissions before usingagy_execwith edit-capable modes.Directory paths are supplied as process arguments and used as the child working directory.
Model validation rejects non-Gemini model names before starting
agy.Do not commit
agycredentials, local configuration files, or environment secrets.
Project Structure
src/
├── config.ts # Binary resolution and Gemini allowlist
├── index.ts # MCP stdio server entry point
├── runner.ts # agy process lifecycle and timeout handling
└── tools/
├── command.ts # Literal slash commands
├── exec.ts # Generic prompts
├── plan.ts # Technical planning
├── review.ts # Code review
└── validate-tests.ts # Test analysisDevelopment
npm install
npm run typecheck
npm run build
npm startThe server communicates through stdin/stdout using MCP JSON-RPC. Diagnostic messages are written to stderr.
License
This project is licensed under the MIT License. See LICENSE.
Available Tools
5 toolsagy_commandB
Executa literalmente um comando direto do Antigravity CLI, como /usage, sem pedir ao modelo para interpretar ou reformular a instrução.
Inputs: command (required), model, directory.
agy_execC
Executa comandos genéricos ou consultas arbitrárias via Antigravity (agy CLI) com suporte a modelos Google Gemini.
Inputs: prompt (required), mode, model, directory.
agy_planC
Gera ou valida planos de arquitetura e implementação técnica utilizando o Antigravity (agy CLI) em modo de planejamento.
Inputs: prompt (required), model, directory.
agy_reviewB
Realiza revisão técnica aprofundada de código, diffs ou implementações via Antigravity (agy CLI), avaliando manutenibilidade, segurança, performance e possíveis bugs.
Inputs: prompt (required), model, directory.
agy_validate_testsC
Analisa código de teste e de produção via Antigravity (agy CLI) para identificar asserções frágeis, casos de borda não cobertos, vazamento de estado e robustez da suíte.
Inputs: prompt (required), model, directory.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 5 tools
Most tools target distinct tasks: agy_plan for planning, agy_review for code review, and agy_validate_tests for test analysis are clearly separable. The main ambiguity is between agy_exec and agy_command, which both execute commands via the CLI; the descriptions distinguish model-interpreted vs. literal execution, but the boundary is subtle enough to risk occasional misselection.
All tools share the predictable agy_ prefix and snake_case, which helps grouping. However, the suffixes mix conventions: agy_plan and agy_command are noun-like, while agy_validate_tests and agy_review are verb phrases, and agy_exec is an abbreviation, so there is no single consistent verb_noun pattern.
Five tools is a well-scoped set for an Antigravity CLI wrapper, covering planning, review, test validation, generic execution, and direct command invocation without redundant bulk.
The surface covers the main modes of interacting with the agy CLI: planning, code review, test analysis, generic execution, and literal command execution. Minor gaps like explicit model/config listing or session management could exist, but core workflows appear covered.
Maintenance
Related MCP Connectors
Talk to your public-facing AI from any MCP client — Claude, ChatGPT, Cursor, Cline, Windsurf.
Use AI models for chat, image, and video generation from Claude Code and other MCP hosts.
Connect ChatGPT and Claude to your Mac, Windows or Linux with secure MCP remote access.
Official MCP bridge to Exogram—the Personal AI Agent for your work, people, and life.
Related MCP Servers
- AlicenseNot gradedqualityAmaintenanceMCP server enabling delegation of tasks to the Antigravity (Gemini) CLI from any AI client supporting MCP.8 npmMIT
- AlicenseNot gradedqualityDmaintenanceEnables MCP-compatible coding agents to run the local Antigravity CLI as a coding agent, manage conversation context and common options, and inspect usage, quota, models, version, help, and read-only slash commands.26 npmMIT
- AlicenseBqualityBmaintenanceEnables MCP-compatible AI clients like Claude Code, Cursor, and Codex to invoke Google Antigravity's Gemini coding agent for repository-aware file operations, command execution, code review, test generation, and quota checks, while also providing a standalone dual-mode CLI.7MIT
- AlicenseAqualityBmaintenanceEnables MCP clients to use the local Antigravity CLI for natural-language queries, repository analysis, and agent execution, reusing the user's Antigravity subscription without exporting credentials.8MIT