native-mcp

Connects to MCP servers and registers their tools as native agent tools.

14|5|Updated Apr 9, 2026
One-click install
npx skills add https://github.com/MLT-OSS/hermes-agent-go --skill native-mcp-mlt-oss
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: native-mcp
Source: https://github.com/MLT-OSS/hermes-agent-go/tree/main/skills/mcp/native-mcp
Command: npx skills add https://github.com/MLT-OSS/hermes-agent-go --skill native-mcp-mlt-oss

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires mcp.

What problem does it solve? Integrating external MCP (Model Context Protocol) servers into an AI agent normally requires custom bridge code and manual tool wiring. This Skill provides a built-in MCP client that connects to servers at startup, discovers their tools, and registers them as first-class agent tools with zero additional code. ## Core Features & Use Cases - Automatic Tool Discovery: Connects to configured MCP servers on startup, calls list_tools(), and registers each tool with an mcp_{server}{tool} naming prefix. - Dual Transport Support: Runs local stdio servers (npx, uvx, or any command) and remote HTTP/StreamableHTTP servers with custom headers and authentication. - Security Controls: Filters environment variables passed to subprocesses, redacts credentials from error messages, and supports per-server sampling limits. - Use Case: Add a GitHub MCP server with a personal access token to your config, restart the agent, and immediately ask it to list issues or create pull requests using the auto-registered mcp_github* tools. ## Quick Start Add an mcp_servers entry with a command and args to ~/.hermes/config.yaml, then restart the agent and ask it to use the newly discovered 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 my AI agent?

Add an entry under the mcp_servers key in ~/.hermes/config.yaml with either a command and args for stdio servers or a url for HTTP servers. Restart the agent and its tools are discovered and registered automatically with an mcp_{server}_{tool} prefix.

How do I run npx or uvx MCP servers locally?

Set the command field to npx or uvx and pass the package name in args, for example command: npx with args: ["-y", "@modelcontextprotocol/server-filesystem", "/path"]. Node.js is required for npx servers and uv for uvx servers.

Does the MCP client support remote HTTP servers?

Yes, HTTP and StreamableHTTP transports are supported by setting a url instead of a command, with optional headers for authentication. Your installed mcp package must include mcp.client.streamable_http support.

Are my environment variables and API keys safe with MCP servers?

Stdio subprocesses receive only a filtered baseline of safe variables like PATH and HOME, not your full shell environment. Credentials must be explicitly passed via the env config key, and credential patterns are redacted from error messages.

Why are my MCP tools not appearing after configuration?

Check that servers are listed under the mcp_servers key with correct YAML indentation and that the mcp Python package is installed. Tools are prefixed with mcp_{server}_{tool}, so search for that naming pattern in startup logs.

What happens when an MCP server connection drops?

The client automatically reconnects with exponential backoff up to 5 retries, capped at 60 seconds between attempts. If the server remains unreachable after all retries, that server is abandoned while others continue normally.