One-click install
npx skills add https://github.com/JamesFincher/gengar --skill native-mcp-jamesfincher
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: native-mcp
Source: https://github.com/JamesFincher/gengar/tree/main/skills/mcp/native-mcp
Command: npx skills add https://github.com/JamesFincher/gengar --skill native-mcp-jamesfincher

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

It eliminates manual setup and bridging when you want to use external Model Context Protocol (MCP) servers so their tools become immediately available inside Gengar.

Core Features & Use Cases

  • MCP server discovery at startup: Reads your configured MCP servers, connects, and lists available tools automatically.
  • First-class tool registration: Exposes discovered MCP tools alongside built-in tools, prefixed as mcp_{server_name}_{tool_name}, so the agent can call them naturally.
  • Configurable transports (stdio or HTTP): Runs local stdio-based MCP servers via commands (e.g., npx/uvx) or connects to remote HTTP/StreamableHTTP MCP endpoints.
  • Security guardrails for subprocess environments: Filters inherited environment variables for stdio subprocesses and supports explicit env injection only when required.
  • Credential redaction on failures: Automatically redacts common credential-like patterns from tool error messages.

Quick Start

Add your MCP servers to ~/.gengar/config.yaml under mcp_servers, restart Gengar, then ask it to use an MCP tool (for example, “What is the current time?”) with the automatically generated mcp_{server}_{tool} prefix.

Frequently Asked Questions about native-mcp

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I connect an MCP server to my agent for tool discovery?

MCP server tool discovery is handled by connecting to configured servers at startup, automatically listing their tools, and registering them with an mcp_{server_name}_{tool_name} prefix for native agent usage. You add servers to your config file and restart.

Can I use both stdio and HTTP transports for MCP integrations?

Both stdio and HTTP transports are supported for MCP integrations. You can run local stdio-based MCP servers via commands like npx or uvx, and also connect to remote HTTP/StreamableHTTP MCP endpoints during the recurring startup initialization flow.

How are credentials protected when running stdio MCP servers?

Credentials are protected during stdio MCP server execution by filtering inherited environment variables and only allowing explicit environment injection when required. Common credential-like patterns are also automatically redacted from tool error messages.

Do I need the mcp Python package to use MCP tool discovery?

The mcp Python package is required for optional support to enable MCP tool discovery and native registration. This dependency provides the underlying connection management and tool execution capabilities needed to interface with configured servers.

How do I call an MCP tool after it has been discovered?

Discovered MCP tools are called naturally by asking the agent to use them with the automatically generated mcp_{server_name}_{tool_name} prefix. They are exposed alongside built-in tools, making them immediately available for native agent usage.