antigravity-bridge
MCP Server connecting Claude Desktop to Antigravity CLI (agy) with automated git diff code review
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Badge for your README
Markdown
[](/listings/thecrystoria-svg--antigravity-bridge)HTML
<a href="/listings/thecrystoria-svg--antigravity-bridge"><img src="/api/listings/thecrystoria-svg--antigravity-bridge/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
These steps run a script from the network that findsafeskills does not scan.
Option 1: Git Clone (Recommended & Auditable)
Clone the repository to inspect all local automation scripts (`install.py`, `agy_bridge.py`) prior to launch:
git clone https://github.com/thecrystoria-svg/antigravity-bridge.git cd antigravity-bridge python install.py
- **Windows**: Run `setup.bat` or `python install.py`. - **macOS / Linux**: Run `./setup.sh` or `python3 install.py`.
The installer checks your environment, prepares an isolated virtual environment, configures `claude_desktop_config.json`, adjusts headless permissions, and executes 9 self-verification tests automatically.
Once installed, restart **Claude Desktop**.
---
Option 2: 1-Line Web Installer
- **Windows PowerShell**:
irm https://raw.githubusercontent.com/thecrystoria-svg/antigravity-bridge/main/install.ps1 | iex
- **macOS / Linux Shell**:
curl -fsSL https://raw.githubusercontent.com/thecrystoria-svg/antigravity-bridge/main/install.sh | bash
---
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
README.md · README · 356 lines · 2 findings marked
- # Antigravity Claude Bridge
- > **Autonomous Multi-Agent MCP Server connecting Claude Desktop to the Antigravity CLI (`agy`).**
- [](https://www.python.org/)
- [](https://modelcontextprotocol.io/)
- []()
- [](LICENSE)
- []()
- The **Antigravity Claude Bridge** is a local Model Context Protocol (MCP) server that empowers **Claude Desktop** to act as a **Chief Systems Architect & CEO**, delegating end-to-end coding, directory scaffolding, test execution, and diff generation to **Antigravity CLI (`agy`)** as its autonomous engineering engine.
- ---
- ## Architecture Overview
- ```
- ┌────────────────────────────────────────────────────────┐
- │ Claude Desktop (CEO / Architect) │
- │ - Formulates system designs, schemas & milestones │
- │ - Delegates engineering tracks directly to Antigravity │
- │ - Inspects file implementations line-by-line │
- │ - Reviews terminal test logs, diffs & exit codes │
- └───────────────────────────┬────────────────────────────┘
- │ (MCP stdio protocol)
- ▼
- ┌────────────────────────────────────────────────────────┐
- │ agy_bridge.py (FastMCP Server) │
- │ - Embedded CEO Protocol & Execution Tool Suite │
- │ - Automated Git & non-Git snapshot diff engine │
- └───────────────────────────┬────────────────────────────┘
- │ (--dangerously-skip-permissions + wildcard allow)
- ▼
- ┌────────────────────────────────────────────────────────┐
- │ Antigravity CLI Swarm (Chief Engineer) │
- │ - Creates folders & scaffolds directory trees │
- │ - Authors modular, zero-regression source files │
- │ - Executes automated tests & captures stdout/stderr │
- │ - Delivers structured milestone summaries & diffs │
- └────────────────────────────────────────────────────────┘
- ```
- ---
- ## Why Antigravity Bridge?
- | Capability | Generic MCP Bridges | Antigravity Claude Bridge |
- | :--- | :--- | :--- |
- | **Permission Friction** | Prompts for terminal approval on every directory | **Zero prompts**: Wildcard allow + auto-bypass flag |
- | **Workspace Support** | Fails or returns empty diffs on non-Git folders | **Universal**: Native Git diffs or recursive snapshot diffs |
- | **Agent Roles** | Single generic execution call | **Specialized subagents**: Architect, Coder, Reviewer, Tester |
- | **Code Verification** | Blind output without proof of test results | **Live evidence**: Captures test commands, exit codes & logs |
- | **Setup Experience** | Manual JSON edits and missing PATH errors | **1-Click installer**: Environment doctor + automated self-tests |
- ---
- ## Quickstart
- ### Option 1: Git Clone (Recommended & Auditable)
- Clone the repository to inspect all local automation scripts (`install.py`, `agy_bridge.py`) prior to launch:
- ```bash
- git clone https://github.com/thecrystoria-svg/antigravity-bridge.git
- cd antigravity-bridge
- python install.py
- ```
- - **Windows**: Run `setup.bat` or `python install.py`.
- - **macOS / Linux**: Run `./setup.sh` or `python3 install.py`.
- The installer checks your environment, prepares an isolated virtual environment, configures `claude_desktop_config.json`, adjusts headless permissions, and executes 9 self-verification tests automatically.
- Once installed, restart **Claude Desktop**.
- ---
- ### Option 2: 1-Line Web Installer
- - **Windows PowerShell**:
- ```powershell
- irm https://raw.githubusercontent.com/thecrystoria-svg/antigravity-bridge/main/install.ps1 | iex
- Note: Runs a downloaded PowerShell script (iex / Invoke-Expression)
- Line 82 on GitHub
- ```
- - **macOS / Linux Shell**:
- ```bash
- curl -fsSL https://raw.githubusercontent.com/thecrystoria-svg/antigravity-bridge/main/install.sh | bash
- Note: Pipes a downloaded script directly into a shell (curl | sh / wget | bash)
- Line 87 on GitHub
- ```
- ---
- ## Available MCP Tools
- Claude Desktop gains access to 17 dedicated engineering tools:
- ### 1. `antigravity_list_models`
- Dynamically queries the live list of models and reasoning levels available from the Antigravity CLI (`agy models`). Categorizes models into **Ultra-Fast** (10–25s response time), **Balanced**, and **Deep Reasoning / High-Thinking** tiers. Enables Claude to discover newly released models automatically and ask the user for their speed preference.
- ### 2. `antigravity_ceo_charter`
- CEO Master Governance Tool: Establishes the authoritative project charter, assigns team roster roles, specifies the modular code architecture (Clean Architecture / Domain-Driven Design), defines token-saving invariants, and commits `PROJECT_CHARTER.md` to disk.
- | Parameter | Type | Required | Description |
- | :--- | :--- | :--- | :--- |
- | `workspace_path` | `string` | Yes | Target project workspace directory. |
- | `project_name` | `string` | Yes | Name of the project or application. |
- | `mission` | `string` | Yes | High-level business objective, requirements, and deliverables. |
- | `tech_stack` | `string` | No | Technologies: e.g. `'FastAPI + PostgreSQL + Next.js'`. |
- | `architecture_pattern` | `string` | No | Architecture pattern: e.g. `'Clean Architecture (Core, Interfaces, Services, Repositories, Controllers)'`. |
- | `team_roster` | `string` | No | Subagent roles mobilized: e.g. `'PM, Architect, Backend Coder, Frontend Coder, QA, CTO, SRE, Inspector'`. |
- | `token_strategy` | `string` | No | Token-friendly rules: micro-task decomposition, targeted file inspection, diff-only reporting. |
- | `model` | `string` | No | Optional model override. |
- | `effort` | `string` | No | Reasoning effort: `'low'`, `'medium'`, `'high'`. Defaults to `'high'`. |
- ### 3. `antigravity_ceo_delegate`
- High-level executive orchestration: Claude commands Antigravity across designated engineering phases (`plan`, `build`, `test`, `review`).
- | Parameter | Type | Required | Description |
- | :--- | :--- | :--- | :--- |
- | `directive` | `string` | Yes | High-level directive or user requirements from Claude. |
- | `workspace_path` | `string` | Yes | Target project directory path. |
- | `phase` | `string` | No | Target phase: `'plan'`, `'build'`, `'test'`, `'review'`, `'auto'`. Defaults to `'plan'`. |
- | `context` | `string` | No | Architectural rules, schema constraints, or style guides. |
- | `model` | `string` | No | Optional LLM override (e.g. `gemini-3.8-flash-low`, `claude-sonnet-4-6`). |
- | `effort` | `string` | No | Reasoning effort: `'low'`, `'medium'`, `'high'`, `'xhigh'`, `'max'`. Defaults to `'high'`. |
- ### 4. `antigravity_cto_security_audit`
- Chief Technology Officer (Security & Tech Stack) tool: Audits code against OWASP Top 10 vulnerabilities (SQLi, XSS, CSRF, IDOR), scans package manifests for known CVEs and outdated dependencies, enforces zero-trust secret protection, and generates `SECURITY_AUDIT.md`.
- | Parameter | Type | Required | Description |
- | :--- | :--- | :--- | :--- |
- | `workspace_path` | `string` | Yes | Target project workspace directory. |
- | `auto_patch_vulnerabilities` | `boolean` | No | If True, updates vulnerable packages to safe patched versions and applies fixes. |
- | `context` | `string` | No | Security compliance targets (SOC2, HIPAA, PCI-DSS) or threat model context. |
- | `model` | `string` | No | Optional model override. |
- | `effort` | `string` | No | Reasoning effort: `'low'`, `'medium'`, `'high'`. Defaults to `'high'`. |
- ### 5. `antigravity_cfo_audit`
- Chief Financial Officer (FinOps) tool: Audits package dependencies, software licensing liabilities (MIT, Apache vs AGPL/GPL copyleft), cloud runtime compute tier, API token costs, and generates `FINOPS.md`.
- | Parameter | Type | Required | Description |
- | :--- | :--- | :--- | :--- |
- | `workspace_path` | `string` | Yes | Target project workspace directory. |
- | `context` | `string` | No | Cost targets, cloud provider preferences, or monthly active user estimates. |
- | `model` | `string` | No | Optional model override. |
- | `effort` | `string` | No | Reasoning effort: `'low'`, `'medium'`, `'high'`. Defaults to `'high'`. |
- ### 6. `antigravity_coo_audit`
- Chief Operating Officer (DevOps) tool: Audits operational readiness, CI/CD pipelines (`.github/workflows`), Docker containerization, health-check endpoints, secret isolation, and generates `RUNBOOK.md`.
- | Parameter | Type | Required | Description |
- | :--- | :--- | :--- | :--- |
- | `workspace_path` | `string` | Yes | Target project workspace directory. |
- | `context` | `string` | No | Target deployment platform (AWS, GCP, Cloudflare, Docker) or uptime SLA. |
- | `model` | `string` | No | Optional model override. |
- | `effort` | `string` | No | Reasoning effort: `'low'`, `'medium'`, `'high'`. Defaults to `'high'`. |
- ### 7. `antigravity_deployment_scaffold`
- Principal Deployment Engineer & SRE tool: Scaffolds production-grade deployment configurations (multi-stage Dockerfile, docker-compose.yml, Kubernetes manifests, Cloud Run / Fly.io configs, zero-downtime scripts) and generates `DEPLOYMENT.md`.
- | Parameter | Type | Required | Description |
- | :--- | :--- | :--- | :--- |
- | `workspace_path` | `string` | Yes | Target project workspace directory. |
- | `target_platform` | `string` | No | Environment: `'docker'`, `'kubernetes'`, `'cloud-run'`, `'fly.io'`, `'serverless'`, `'all'`. Defaults to `'docker'`. |
- | `context` | `string` | No | Domain names, port mappings, environment secrets, or database URLs. |
- | `model` | `string` | No | Optional model override. |
- | `effort` | `string` | No | Reasoning effort: `'low'`, `'medium'`, `'high'`. Defaults to `'high'`. |
- ### 8. `antigravity_decompose_tasks`
- Project Manager tool: Breaks down a complex user objective into 3 to 7 small, sequential micro-tasks and writes `TASKS.md` in the workspace root.
- | Parameter | Type | Required | Description |
- | :--- | :--- | :--- | :--- |
- | `objective` | `string` | Yes | High-level feature request or engineering goal to decompose. |
- | `workspace_path` | `string` | Yes | Target project workspace directory. |
- | `context` | `string` | No | Architectural constraints, tech stack preferences, or invariants. |
- | `model` | `string` | No | Optional model override. |
- | `effort` | `string` | No | Reasoning effort: `'low'`, `'medium'`, `'high'`. Defaults to `'high'`. |
- ### 9. `antigravity_execute_task`
- Autonomous execution of a single atomic micro-task from `TASKS.md` (e.g. `'Task 1'`). Implements code, runs tests, updates `TASKS.md` to completed `[x]`, and generates a discrete Git commit. Defaults to `effort='high'` for deep reasoning and zero-defect implementations.
- | Parameter | Type | Required | Description |
- | :--- | :--- | :--- | :--- |
- | `task_id` | `string` | Yes | Identifier of the task to execute (e.g. `'Task 1'`, `'Task 2'`). |
- | `workspace_path` | `string` | Yes | Target project workspace directory. |
- | `task_instructions` | `string` | No | Specific implementation guidance or code requirements. |
- | `model` | `string` | No | Optional model override (e.g. `'gemini-3.8-flash-high'`). |
- | `effort` | `string` | No | Reasoning effort: `'low'`, `'medium'`, `'high'`, `'xhigh'`, `'max'`. Defaults to `'high'`. |
- ### 10. `antigravity_inspect_resume`
- Inspector & Recovery tool: Audits the workspace after an error, interruption, network drop, or partial edit. Checks `TASKS.md` progress (`[x]` vs `[ ]`), inspects git status for partially written files, runs test verification, authors `INSPECTOR_RECOVERY.md`, and automatically resumes execution from the exact point of interruption without duplicating work.
- | Parameter | Type | Required | Description |
- | :--- | :--- | :--- | :--- |
- | `workspace_path` | `string` | Yes | Target project workspace directory. |
- | `auto_resume_next_task` | `boolean` | No | If True, automatically launches execution of the next incomplete task. Defaults to `true`. |
- | `context` | `string` | No | Specific recovery guidance or instructions. |
- | `model` | `string` | No | Optional model override. |
- | `effort` | `string` | No | Reasoning effort: `'low'`, `'medium'`, `'high'`. Defaults to `'high'`. |
- ### 11. `run_antigravity_subagent`
- Spins up a specialized autonomous Antigravity subagent and streams milestone progress, technical decisions, and diffs to Claude.
- | Parameter | Type | Required | Description |
- | :--- | :--- | :--- | :--- |
- | `task_description` | `string` | Yes | Detailed user story or task prompt. |
- | `role` | `string` | No | Persona: `'inspector'`, `'cto'`, `'deployment-engineer'`, `'cfo'`, `'coo'`, `'pm'`, `'researcher'`, `'backend-coder'`, `'frontend-coder'`, `'coder'`, `'qa'`, `'architect'`, `'code-reviewer'`, `'tester'`, `'debugger'`. Defaults to `'coder'`. |
- | `workspace_path` | `string` | Yes | Target project directory path. |
- | `context` | `string` | No | Contextual background or system invariants. |
- | `model` | `string` | No | Optional model override. |
- | `effort` | `string` | No | Optional reasoning effort. Defaults to `'high'`. |
- | `auto_init_git` | `boolean` | No | If True and workspace is not a Git repo, runs `git init`. Defaults to `true`. |
- ### 12. `antigravity_get_history`
- Returns the chronological history of all phases, directives, git commits, and file modifications in the workspace. Enables Claude to maintain full context on large projects without re-reading entire codebases.
- | Parameter | Type | Required | Description |
- | :--- | :--- | :--- | :--- |
- | `workspace_path` | `string` | Yes | Target workspace directory path. |
- | `limit` | `integer` | No | Number of recent milestones to return. Defaults to `10`. |
- ### 13. `antigravity_read_file`
- Enables Claude to inspect specific source files line-by-line within the workspace to perform deep code reviews before approving changes.
- | Parameter | Type | Required | Description |
- | :--- | :--- | :--- | :--- |
- | `workspace_path` | `string` | Yes | Target workspace directory path. |
- | `relative_file_path` | `string` | Yes | Relative file path (e.g. `src/server.py`, `tests/test_api.py`). |
- ### 14. `antigravity_run_tests`
- Executes test suites in the target project and returns exact terminal output, pass/fail status, and exit codes for Claude to audit.
- | Parameter | Type | Required | Description |
- | :--- | :--- | :--- | :--- |
- | `workspace_path` | `string` | Yes | Target workspace directory path. |
- | `test_command` | `string` | No | Custom command (e.g. `pytest`, `npm test`, `cargo test`). If omitted, auto-detects based on project files. |
- ### 15. `run_antigravity_task`
- General-purpose task delegation to Antigravity CLI returning structured status and git diffs.
- | Parameter | Type | Required | Description |
- | :--- | :--- | :--- | :--- |
- | `prompt` | `string` | Yes | Prompt or instruction string for Antigravity. |
- | `workspace_path` | `string` | Yes | Target workspace directory path. |
- | `model` | `string` | No | Optional model override. |
- | `effort` | `string` | No | Reasoning effort (`'low'`, `'medium'`, `'high'`). Defaults to `'high'`. |
- | `agent` | `string` | No | Optional specialized agent profile name. |
- | `auto_accept_edits` | `boolean` | No | Automatically accept code edits (`--mode accept-edits`). Defaults to `true`. |
- | `auto_init_git` | `boolean` | No | Automatically initialize git tracking if absent. Defaults to `true`. |
- ### 16. `get_workspace_diff`
- Fetches unstaged, staged, and untracked file diffs (or recursive directory snapshot changes if non-Git) without initiating a new generation task.
- | Parameter | Type | Required | Description |
- | :--- | :--- | :--- | :--- |
- | `workspace_path` | `string` | Yes | Target workspace directory path. |
- ### 17. `check_antigravity_status`
- Reports active CLI path, installed version, Google authentication status, and configured permission flags.
- ---
- ## Example Prompts in Claude Desktop
- ### Example 1: Full-Stack Feature Delegation (CEO Mode)
- > *"You are the CEO and Principal Architect. Analyze this requirement: build an SQLite-backed URL shortener with FastAPI and Pytest in `C:\projects\url-shortener`. Delegate folder setup, implementation, and test suite creation to Antigravity, then audit the code and test logs."*
- ### Example 2: Specialized Subagent Execution
- > *"Launch an Antigravity 'architect' subagent in `C:\projects\my-app` to define the database schema and endpoints for user authentication. Then, trigger the 'coder' subagent to implement the models and return the diff."*
- ### Example 3: Pre-Merge Code Audit
- > *"Run an Antigravity 'code-reviewer' subagent in `C:\projects\my-app` to inspect recent commits for security vulnerabilities, injection risks, and unhandled edge cases."*
- ---
- ## Large Projects & Multi-Turn History Tracking
- Building extensive codebases requires resilience across dozens of turns without token waste or loss of context:
- 1. **Continuous Conversation Memory (`--conversation`)**:
- Antigravity records a session token per workspace in `.antigravity/session.json`. Subsequent tool calls resume the exact conversation state, preserving technical decisions, user instructions, and prior file edits across turns.
- 2. **Persistent Track History (`TRACK_HISTORY.md`)**:
- Every delegation automatically appends a human-readable milestone entry and an append-only JSONL record (`.antigravity/history.jsonl`), recording the phase, directive, git commit, touched files, and summary.
- 3. **Automated Git Checkpointing**:
- After each completed phase, changes are committed to Git with structured messages (e.g. `antigravity: [BUILD] Implement authentication endpoints`), ensuring you can inspect or revert any step with standard `git diff` or `git reset`.
- 4. **Concurrency Safety (`WorkspaceLock`)**:
- Prevents file write collisions or duplicate processes when running rapid subagents or parallel requests.
- 5. **High-Performance Large Codebase Traversal**:
- Automatically ignores heavy build directories (`node_modules`, `venv`, `dist`, `build`, `target`, `.next`, `coverage`) to remain fast and memory-light even on 100,000+ file monorepos.
- ---
- ## Running Verification Tests
- To verify your local bridge installation:
- ```bash
- # Using system Python:
- python -m unittest
- # Or via the dedicated virtual environment:
- # Windows:
- .\.venv\Scripts\python.exe -m unittest discover -s tests
- # macOS / Linux:
- ./.venv/bin/python -m unittest discover -s tests
- ```
- ---
- ## Frequently Asked Questions (FAQ)
- #### Q: How does this prevent Antigravity permission dialogs?
- The bridge configures `~/.gemini/antigravity-cli/settings.json` with wildcard allow rules (`["command(*)", "write_file(*)", "edit_file(*)", "read_file(*)", "*"]`) and launches `agy` with `--dangerously-skip-permissions`.
- #### Q: What happens if my project folder does not have Git initialized?
- The bridge defaults to `auto_init_git=True` to initialize Git tracking automatically. If Git is not installed, the bridge uses an internal pre/post snapshot diff engine to capture all new and modified files.
- #### Q: Where are Claude Desktop configuration files stored?
- - **Windows**: `%APPDATA%\Claude\claude_desktop_config.json`
- - **macOS**: `~/Library/Application Support/Claude/claude_desktop_config.json`
- - **Linux**: `~/.config/Claude/claude_desktop_config.json`
- ---
- ## Dynamic Model Discovery & Automatic Fallback Engine
- ### 1. Live Model Discovery
- The bridge executes `agy models` dynamically in real-time. Whenever Google, Anthropic, or Antigravity adds new models (e.g., future Gemini or Claude versions) or when your account gains access to new model tiers, they **automatically appear in `antigravity_list_models`** with zero manual updates required.
- ### 2. Automatic Quota & Token Exhaustion Fallback
- If any model encounters rate limits (429), quota exhaustion, or capacity errors:
- 1. The bridge automatically detects the error and switches to the next model in the fallback chain:
- `gemini-3.8-flash-high` ➔ `gemini-3.8-flash-medium` ➔ `gemini-3.8-flash-low` ➔ `claude-sonnet-4-6` ➔ `claude-opus-4-6-thinking` ➔ `gpt-oss-120b-medium`.
- 2. If all Antigravity CLI models are exhausted, the bridge returns a structured handover instructing Claude Desktop (CEO) to complete the task directly with Claude's native tools.
- ### 3. Lead Quality & Recovery Inspector (`antigravity_inspect_resume`)
- If an unexpected crash, network drop, or partial write occurs:
- - Run `antigravity_inspect_resume` or launch subagent `role='inspector'`.
- - The Inspector scans `TASKS.md`, identifies completed `[x]` vs pending `[ ]` tasks, checks `git status` for uncommitted in-flight files, verifies test health, writes `INSPECTOR_RECOVERY.md`, and resumes work seamlessly from the exact point of interruption without duplicating code or losing work.
- ---
- ## Disclaimer & Caution
- > [!CAUTION]
- > **Use at Your Own Discretion / Limitation of Liability**
- > - **Zero-Warranty / "AS IS"**: This software and autonomous agents are provided "AS IS", without warranty of any kind, express or implied, including but not limited to the warranties of merchantability, fitness for a particular purpose, and non-infringement.
- > - **Code & Dependency Verification**: While the bridge provides automated CTO security scans, OWASP audits, dependency CVE checks, and automated unit testing, you should always inspect generated code, database migrations, and cloud deployment configs before running in production environments.
- > - **Token & Account Quotas**: Autonomous execution consumes model tokens under your authenticated accounts (Google Antigravity CLI / Anthropic). Monitor your account usage to prevent unexpected quota exhaustion.
- ---
- ## License
- This project is licensed under the [MIT License](LICENSE).
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.