16_mcp_connector_skill

Overview

This skill exposes the entire Antigravity skills library as an MCP (Model Context Protocol) server so MCP-compliant agents can call skills as native tools. It acts as a meta-skill wrapper, translating MCP requests to the underlying PowerShell-based skill implementations. The server is lightweight and intended to run as a persistent service that agents connect to over stdio.

How this skill works

The MCP Connector launches a Python-based MCP server that registers available skills and their specs, then exposes a small set of tool calls (list_skills, run_skill, read_skill_spec) over stdio. Incoming MCP calls are validated, mapped to the PowerShell skill implementations, and executed with deterministic execution and safety checks enforced by the library. The server requires Python 3.10+ and the mcp package to operate and is configured as an MCP server entry for external clients.

When to use it

• You want to let Claude Desktop or any MCP-compliant agent use Antigravity skills without embedding PowerShell logic in the agent.
• You need a stable, tested bridge between an external agent and the skill library for production workflows.
• You want centralized enforcement of safety gates, impasse detection, and failure postmortems for agent-invoked actions.
• You need to expose skill metadata and run skills remotely via tool calls over stdio.
• You want to run deterministic, auditable skill executions from third-party agents.

Best practices

• Run the server in a controlled environment where the PowerShell runtime and required packages are managed.
• Ensure Python 3.10+ and the mcp package are installed and pinned to known versions for reproducibility.
• Register the server in your agent config (e.g., claude_desktop_config.json) so clients discover it via MCP.
• Limit server permissions and isolate execution contexts to minimize risk from invoked skills.
• Monitor logs and use the library’s safety gates and postmortem features to detect and remediate issues.

Example use cases

• Enable Claude Desktop to list and invoke Antigravity skills as if they were built-in tools.
• Integrate the skills library into a multi-agent pipeline where one agent delegates tasks to Antigravity.
• Provide a deterministic execution backend for CI tests that exercise skill behavior via MCP calls.
• Expose skill specs to a user interface that dynamically builds tool palettes based on available skills.
• Automate high-leverage developer workflows by having an external agent orchestrate skill runs.