Skip to main content
Glama

🚀 MetaGravity (metagravity-mcp)

License Python MCP

MetaGravity is a Model Context Protocol (MCP) Server and Python SDK for orchestrating Google Antigravity agents and CLI sessions. Inspired by T3 Code's autonomous multi-chat spawning, MetaGravity enables one lead chat or script to spawn, steer, and archive multiple child agent sessions across GitHub issues or feature folders.


💡 Why MetaGravity?

In harnesses like T3 Code, an agent can effortlessly say:

"Open 3 parallel chats in this repo, one for each of these 3 GitHub issues, execute them in background, and give them initial prompts."

Antigravity natively supports reading and continuing threads via agy, but lacked an effortless, standardized MCP interface for programmatic multi-chat dispatch. MetaGravity bridges that gap.

Core Primitives

  1. launch_chat: Spawns a dedicated Antigravity session in target project folder X, with configurable permissions (turbo / --dangerously-skip-permissions, auto, plan), and an initial trigger prompt.

  2. add_comment: Injects follow-up prompts or answers interactive choices (e.g., auto-selecting option 4 when a session is paused waiting for user confirmation).

  3. archive_chat: Formally archives / tombstones completed conversations in Antigravity's metadata (annotations/<cid>.pbtxt) so active monitoring stays clean.

  4. list_chats: Discovers active and past threads directly from Antigravity's internal SQLite database (conversation_summaries.db).

  5. get_transcript: Reads step-by-step logs and tool outputs from Antigravity's brain transcript.


Related MCP server: switchboard

🛠️ Quickstart

1. Installation

git clone https://github.com/palladius/metagravity-mcp.git
cd metagravity-mcp
pip install -e .

Or run directly with uv:

uv run --project . metagravity mcp

2. Configure with your MCP Client (Claude Desktop, Cursor, Hermes, T3)

Add to your claude_desktop_config.json or MCP settings:

{
  "mcpServers": {
    "metagravity": {
      "command": "python3",
      "args": ["-m", "metagravity.server"],
      "cwd": "/path/to/metagravity-mcp"
    }
  }
}

🧰 Available MCP Tools

Tool

Parameters

Description

metagravity_launch_chat

project, prompt, mode (auto), turbo (true), background (true)

Spawns a new Antigravity agent in the project directory.

metagravity_add_comment

conversation_id, comment, choice (optional int)

Appends a comment or resolves a 1-2-3-4 selection prompt.

metagravity_archive_chat

conversation_id

Tombstones and archives a conversation.

metagravity_list_chats

limit (int, default 20), open_only (bool)

Lists active/past threads with status and step count.

metagravity_get_transcript

conversation_id, limit (int, default 30)

Inspects recent tool calls and steps from the agent's brain.


💻 CLI Usage

MetaGravity also doubles as a standalone command-line manager:

# List active chats
metagravity list --open

# Launch a new autonomous chat on a GitHub issue
metagravity launch --project ~/git/my-repo --prompt "Fix issue #12: add healthcheck endpoint" --turbo

# Answer option 4 on a waiting chat
metagravity comment 96b7e2e7 --choice 4

# Archive a finished conversation
metagravity archive 96b7e2e7

🐍 Python SDK Example

from metagravity import MetaGravityClient

client = MetaGravityClient()

# Launch 3 parallel agent chats for 3 GitHub issues
issues = [
    (101, "Fix memory leak in cache poller"),
    (102, "Add dark mode toggle to UI"),
    (103, "Update integration test suite"),
]

sessions = []
for issue_id, title in issues:
    res = client.launch_chat(
        project="/Users/riccardo/git/my-project",
        prompt=f"Please solve GitHub Issue #{issue_id}: {title}. Run tests when done.",
        turbo=True,
        background=True,
    )
    print(f"Spawned Issue #{issue_id} -> PID: {res['pid']}")
    sessions.append(res)

📄 License

Apache-2.0 - Built with 💛 by Riccardo Carlesso & Ermete Bottazzi.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      C
      maintenance
      Enables multiple AI coding CLIs (Claude Code, Gemini/Antigravity, Codex, and OpenCode) to collaborate as a coordinated team by routing cross-agent prompts, sharing messages and review tickets, tracking tasks on a shared store, and isolating each agent in its own Git worktree with turn-budget safeguards—all inspectable and steerable from a local web dashboard.
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables Codex users to delegate engineering, refactoring, auditing, and test execution tasks to Google Antigravity CLI as a persistent autonomous subagent, with tools for managing sessions and running prompts.
      1
      MIT