native-mcp

Connect MCP servers via stdio or HTTP and register their tools as agent tools.

1|Updated Jun 21, 2026
One-click install
npx skills add https://github.com/tapway/shogun-os --skill native-mcp-tapway
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: native-mcp
Source: https://github.com/tapway/shogun-os/tree/main/skills/hermes/mcp/native-mcp
Command: npx skills add https://github.com/tapway/shogun-os --skill native-mcp-tapway

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires mcp, and includes references (resource) components.

What problem does it solve? Configuring external capabilities for an AI agent normally requires custom bridge code or manual tool wiring. This Skill lets you declare MCP servers in a YAML config and have their tools automatically discovered, named, and injected into every conversation at startup. ## Core Features & Use Cases - Dual Transport Support: Connect local stdio servers (npx/uvx commands) or remote HTTP/StreamableHTTP servers with headers and timeouts. - Automatic Tool Discovery & Naming: Tools are registered as mcp_{server}_{tool} and injected into all platform toolsets with reconnection and exponential backoff. - Security Controls: Filtered environment inheritance, explicit env credential passing, and automatic credential stripping from error messages. - Use Case: Add a GitHub MCP server with a personal access token in ~/.hermes/config.yaml, restart the agent, and immediately call tools like mcp_github_list_issues in any conversation. ## Quick Start Add an MCP server entry under mcp_servers in ~/.hermes/config.yaml with a command or url, then restart the agent and ask it to use the newly registered tools.

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 Hermes Agent?

Add an entry under mcp_servers in ~/.hermes/config.yaml with either a command and args for stdio transport or a url for HTTP transport, then restart the agent. Tools are discovered at startup and registered with the mcp_{server}_{tool} naming pattern.

What is the difference between native MCP and mcporter?

The native MCP client is config-driven and auto-discovers tools into every conversation at startup. mcporter is for ad-hoc, one-off MCP tool calls from the terminal without configuring anything; both can be used simultaneously.

Does the MCP client support remote HTTP servers?

Yes, HTTP and StreamableHTTP transports are supported by setting a url and optional headers in the server config. Your installed mcp package must include mcp.client.streamable_http, otherwise upgrade with pip install --upgrade mcp.

Why does my npx-based MCP server fail with Connection closed?

The gateway process often lacks your interactive shell's PATH, so npx or the package binary cannot be found. Install the package globally with npm install -g and use the absolute path to its binary as the command in config.yaml.

Are my environment variables passed to MCP server subprocesses?

No, only safe baseline variables like PATH, HOME, and XDG_* are inherited by stdio subprocesses. API keys and secrets must be explicitly declared in the server's env config block, and credentials are redacted from error messages.

Can MCP servers request LLM completions during tool execution?

Yes, sampling via sampling/createMessage is enabled by default and configurable per server with model overrides, token caps, rate limits, and tool round limits. Disable it for untrusted servers with sampling: { enabled: false }.